3:05 - unpopular opinion - if you were to simply ask me 'which photo do you prefer', I actually would say the R6. Based on what I notice first and foremost, and based on my own priorities, the colours are more saturated and accurate, the contrast is noticeably better, the grain looks consistent and natural, the image looks less processed, and just noticeably crisper edges. The R5 image on the right is more detailed - granted - but my eye is far quicker drawn to the smeary look of Lightroom's noise reduction and the lower contrast, more processed look. I'm a happy owner of an R5 and an R6, but the reason I reach for the R6 more readily for low light work, family dinners etc, is that the files are simply easier to look good with minimal work. I think some of it comes down to how much I dislike the look of Lightroom's luminance noise reduction. I am a big fan of Topaz DeNoise - which I use for wildlife photography but only on special images I intend to print or publish somewhere larger than social media - however the workflow simply isn't fast or convenient enough for bulk editing hundreds of photos from a party or other low light event.
I feel like one of the bigger reasons why people complain about high megapixel noise is on high MP sensors people feel very inclined to crop into images when they normally would not, which magnifies a lot of noise.
This is accurate at least for me. I got an A7Rii for the low light performance and learned this lesson the hard way. You can crop in for days but you lose low light performance.
There is ONE exception to this. The only reason why Sony can put the second "high" 0db preamp at 12,800 ISO on the S-III is because those photosites are a very large 8+ microns in size. You wont ever see secondary preamps programed THAT high on photosites that are 4, 3 or 2 microns. You will see dual native ISO preamps on those smaller photosites...but they will be set MUCH closer together. (Like 800 and 2500 ISO or even closer than that) This is the main ability of that S-III sensor with giant photosites.
That’s not the entire story. ISO12,800 on my FX3 is clean, but must be overexposed 1 to 2 stops, meaning it’s an equivalent ISO3200 from a different camera. ISO 25,600 is clean, as is ISO 51,200 - this is where the true strength of the dual stage of the Sony “S” cameras and (SLOG3). My A7R5 and the A7IV takes better sharper/more detailed video in the day time (And for the A7IV, basically it’s equally strong at night, while my A7R5 only performs as well in 4K30 APS-C Oversampled) than my FX3, especially if I shot in SLOG3 at the equivalent ISOs (Native 2500/800 in my A7R5). When I’m going out in very low conditions like a night market, concert etc where I would want ISO12,800 to ISO51,200, then that’s where I will definitely be taking my FX3 along.
The A7S III uses a nominal 48MP sensor binned down to 12MP, which unfortunately breaks your explanation. It is speculated that the A7S III is set up like this because each sub unit in the 4x4 "pixel" is tuned to a different sensitivity which, with some extra mojo, allows them to produce more dynamic range at the cost of some noise. The A7S II does use large pixels however. You can find this information online, the sensor was x-rayed (which is how this quirk was discovered) and it made the rounds a little while back.
@@perilthecat It's more complex than that. If it's actually 48MP, then how is the A7S III readout speed so much faster than any comparable 48MP camera? This is evidenced by the A7S IIIs excellent rolling shutter. Pixel binning could allow it, but then the noise performance would be terrible. Yet the noise on it, and the FX6 that shares the same sensor is extremely good. Agree that Sony is using a novel technology, but the quantifiable evidence of the camera's output disproves the theory that it's binning, or has a traditional 48MP sensor. Perhaps a fast, on-sensor subsample? Rapidly reading every pixel, discarding no light, and providing a lower resolution sample to the imaging pipeline? Whatever the answer, it's not a simple one.
I feel like most of this is about resolution only here. Not low light high iso noise. You can read Paris and Barcelona at 400% because the a7riv has 5 times the resolution of the a7siii. I'd like to see real pictures at a low light event. And see which are more usable than others.
Those are real pictures aren't they? The point is that with real pictures, you are going to process them, and what you really care about is which gives a better image in the end, no? It's less useful to compare some isolated variable than it is to say 'which sensor provides better final images in low light'.
I agree. Low light capability is much more than end resolution using sliders - you have to consider the dynamic range and colour accuracy - both of wich are not possible to examine by photographing a world map.
Go to Dpreview website and use their Studio scene comparison tool found under Sample Images. There is a Size button, to zoom different megapixel pictures to the same visual size like Tony did. You can compare upto 4 cameras at high iso say 25600, and the scene has gradient bars, detail text, and various elements to compare resolution, dynamic range, color response, etc…. In general Tony is correct. Being a Canon user, I’ve held onto my 2016 50 MP 5DSR for that exact reason. Achievable images on high mpix are not worse, but often sharper with more detail. However, it also depends on how you Denoise. Some programs do a better job than others… so some may smear more detail on the noisier high megapix camera. It is a valid point that the absolute difference between 50 mpix vs say 20 mpix is not leaps and bounds, seen only while pixel peeping. Printed or on a phone, will look similar.
Yep it’s always been the case and that’s why DXO low light (sports) score for high megapixels cameras is pretty much the same as the score for lower megapixel cameras, as DXO cleverly from day 1 id measuring normalized all sensors to 8MP. So 12mp sensor and 45mp sensor gets downscaled to same common number then measured.
Which is accurate for one type of delivery but not all use cases for cameras. Not everyone just shoots to delver 8MP printed photos. One of the main reasons why people use a higher resolution sensor is to crop so they don't need a APSC body or an optical extender which darkens the image by a stop. They like the flexibility to just shoot a photo and crop it as much as they want.
@@digitaldevigner4080 If you're "cropping as much as you want", and you're taking 24mp of the image, then the noise and light performance is the same as a crop sensor or a full frame with a teleconverter. TCs and crop sensor bodies still seem to me to be the much cheaper option given that just 12MP makes for a very high quality image of a normal size, and that's about my only real gripe with the exorbitant price of high MP bodies. There are much cheaper ways to get the same result.
Some high res cams like a7riii, iv and v have better dxo sports rating than some 24mp models. Resolution alone might not account for noise by common sense.
Camera manufactures often don't disclose if the camera is line skipping. Or they line skip in higher demand shooting modes like 4k 60. In addition cameras also have dual gain and perform differently if you're filming at the native iso.
Tweener ISO's are just a push or pull from a native ISO, this applies to stills or vids. But the difference is slight. Bodies like the R5 don't skip lines at full 8K and can shoot RAW. So you get the same quality as in stills mode in vids.
No, you are wrong about this. This is not low light. I specialize in nightlife and live music, videography and photography. For a whole three to four year period I had to bring the a7R3 and the A7S2 with me every time I went out for a job. We're talking no light to some light in scenarios. Now I specifically own the A7S3 and A7IV to still accommodate those scenarios as I had to work a festival on New Year's Eve and then a warehouse after party. The higher megapixel cameras are better photo cameras all together, but they do not handle noise as well and attempting to recover detail or even denoising them will be comparably more difficult for instances where you really have limited amounts of light. They will simply introduce more noise that you cannot reduce as they attempt to find detail in a high ISO scenario. I have been doing this for almost 6 years at arguably the highest levels in this category of content creation. I have endless conversations with my peers comparing our loud outs. I can even tell you which Reds handle best in the same application.
I think many go wrong in the exact way that you show in the video: they zoom to 100% and see more noise. But if you apply denoising at full resolution and then export the image at the same resolution as a lower megapixel camera, you will get slightly better results. The issue is that people like to crop their high mp images which magnifies the noise.
Also easy to fall into the trap of thinking a 400mm will become 600mp because of the high megapixels, but you're also magnifying the noise and the flaws in your lens
The other thing to consider is that with some of the new and amazing noise reduction software we no longer need to make such a trade off of removing noise and losing detail. I would much rather have the raw image with more detail and more noise to start my workflow.
@Phillip Banes That's not true. Due to the nature of Bayer Pattern you lose a lot of potential when reducing the size early stages. Raw denoiser is always gonna be the best denoiser
@Phillip Banes no. You gotta try dxo raw denoiser. It’s free right now for 31 days, no watermarks.. It’s nothing like what we’re used to with denoise sliders in Lr/Ps/etc. Seriously game changing. I no longer feel the need to upgrade my camera body because of noise.
Lol, was it something I said? This comparison is fine for the practical aspect of what it demonstrates, showing that higher resolution sensors capture higher resolution details, but that's as equally true of low ISO levels in bright light. (The A7S III has a 48MP sensor btw, it's binned down to 12MP) Your bucket analogy is incomplete, because it fails to recognize the sources of noise and signal. The pixels of each sensor are not necessarily equal (ignoring size) when it comes to generating a signal from a certain number of photons. This is called quantum efficiency. To borrow a description of the benefits of a BSI vs FSI sensor that just happens to sum it up nicely, from a producer of astrophotography cameras (QHYCCD) where these nitpicks make a big difference: "As a result, more incoming photons strike the photosensitive layer and more electrons are generated and captured in the pixel well. This ratio of photon to electron production is called quantum efficiency. The higher the quantum efficiency the more efficient the sensor is at converting photons to electrons and hence the more sensitive the sensor is to capturing an image of something dim." If you're going to compare the ability to gather signal in low light between different sensors, you need to correct for the sensor's pixels' quantum efficiency. They don't all capture the same amount of rain even if they were the same size, to frame it in your analogy. And your analogy ignores that if each bucket needs 5 rain drops to generate 1 exposure value of signal, then the 9 small buckets will each generate less signal than the 1 large bucket, all other factors being equal. Which means even if you average those 9 buckets the total level of signal that they generate will be less than the 1 bucket spread over the same area. As well, there is the reality of noise and its' sources. To again adapt your analogy, depending on the sensor you are talking about, the one bucket may or may not be introducing the same amount of noise as the nine. Each of the nine could be individually introducing more noise, and thus total noise, or less noise individually but more total noise, or the one could be introducing more noise total than the nine. Noise comes from a variety of areas, both analog and digital. It's a little more complicated to delve into things like input-referred read noise, read noise, analog and digital gain. There's the fact that each camera will have different numbers of ADCs (analog to digital converters) which kick in at different levels. That makes comparing between cameras at the same ISO not so much a test of whether big or small pixels are better for low light but whether that individual camera at that ISO is better than another individual camera at the same ISO. There's also the fact that sensor density best matches itself with optical resolving capability at different focal lengths. Generally speaking, the more telephoto you go the lower the MP sensor you want, as you will otherwise be super-sampling to no benefit in resolution but with a noise penalty (this comes from astronomy where we're using 4000mm focal lengths, mind you). The opposite is true, denser sensors are generally a better match for wider angles. Which is why this video, while demonstrating the slightly (im)practical results of it's simple comparisons, doesn't actually contend with what's going on. And I'm not exactly sure what myth is busted. Bigger pixels - assuming factors such as quantum efficiency are controlled for - are much more capable of generating better SNR. That's a completely different duck from whether high resolution sensors capture more resolution than lower resolution ones.
We may need better empirical analysis to understand more fully the relationship between pixel count and sensor noise for a given line of cameras from a given manufacturer. Tony's brief demonstration seems incomplete to me. It may be subject to experimenter bias and sampling error, as well as a flawed experimental design. We could start by comparing the images produced by the Sony a7R V (61 megapixels) vs the Sony a7 IV (33 megapixels) using a variety of lenses for the Sony E-mount, both entry level and state of the art. We could have professional photographers take hundreds of duplicate images in natural and artificial light, low light, indoors, and outside before examining the results, drawing a conclusion, and moving onto other camera comparisons from Cannon, Nikon, and Fujifilm. Taking this approach would be more scientific, in my opinion, than what Tony did for this video.
@@chesslover8829 I think it would be even more scientific to take cameras apart, scope the sensor analog output, the A/D data, truly measure noise at the analog and at the quantization stages, and do this for various light values. Aggregate results through various binning techniques (probabilistic energy distribution), and come up with envelope error bars that one must now, somehow, objectively, compare between cameras. In other words, go back to the design basics, being privy to the camera manufacturer’s secret pre-amp specifications S/N etc., manufacturing variances…. WOW… not easy.
@billr3053 you don’t need to do any of that actually, Bill Claff of photonstophotos has all the info you could ever want about any camera released in the last decade and a half. Tony has even used some of it when he dipped his toe in the ISO-invariant pool a few years back. Now, just as then, he found a nugget of interest but made a video before looking far enough into it. He missed an opportunity to teach people about how to choose a sensor for their use case, instead choosing to make some faulty generalizations and then run some tests that confirm that higher res sensors produce higher resolution images. Truly groundbreaking
@chesslover8829 You don’t really even need to do any of that, all the relevant information exists already, and any side by side tests like Tony did will be inherently flawed. He’s trying to make a generalization about an oversimplification of how sensors compare. You can’t draw any consistent conclusions from that, because it ignores too many variables in the first place. And he didn’t really have a cohesive “thing” he was trying to show. What does “best” mean? Lowest noise? Most detailed? Most dynamic range? Most sensitive in the shadows? I’m honestly at a loss,
I’ll add an addendum, because it goes unremarked: This “test” completely ignores the spatial filtering that manufacturers such as Sony bake into their RAWs, one purpose of which is to reduce noise in the final image. You can’t turn it off and it’s a different algorithm in every camera (and sometimes changes based on firmware version). To my knowledge the only manufacturer to forgo this was Canon in its DSLRs. This inherently makes a test such as this a fool’s errand, because there are too many uncontrolled variables. And there’s no such thing as “equivalent aperture” for sensor noise and SNR, regardless of format. Imagine taking a photo with a D850, and then changing nothing else you drop it into DX mode and take another photo. What differences will you notice when you compare the two images at the same display resolution? This video is misleading and a missed opportunity
I’m sorry Tony. The Sony A7S series has no equal for dim, exclusively naturally lit, and exclusively hand-held street photography. Just compare the published native ISOs of the various higher MP cameras to the A7 in the comparison. The A7S SEES IN THE DARK… resolving detail not visible to the unaided human eye. I was blown away by my first A7S as it demonstrated better night vision than a Litton image intensifier night scope. If you have any interest in night, hand-held street photography, rent any A7S for yourself and SEE (no pun intended). My Z9 rests comfortably at home when I go night hunting. There is no contest. A7S paired with a superfast wide angle prime (Sigma 20mm f1.4 Art), smiles all night…
Hi Tony.. I have an A7rii and an A7sii. I predominantly shoot astrophotography, and under real low light conditions, the A7s out performs the A7r every time. It's fair to say I shoot entirely in manual and have no requirement to crop or need for focus points, etc. I tend to shoot single shot nightscape exposures, and the A7s will produce far less noise and, as a result, much cleaner images. While I appreciate your points, I beg to differ after years of using both cameras in real-world low light conditions.
I think it is fair to say that for most cameras the video quality still remains better on lower megapixel sensor variants -- not because of the sensor altered features, but because of the limited readout ability with more megapixels!
So if you took HDMI out raw and processed externally where there were less limitations would it be better or equivalent? Or if processors get better and 48 megapixel would give you 8k video, it would be the best? Seems like it is all a processing problem and video has to process, stills are just capturing one image.
@@TonyAndChelsea I am not discrediting you in this regard. You did acknowledge the readout limitations has an impact. And yes, ultimately the comparison has to be apples to apples (ex: A7SIII is far superior vs A7RIV for 4K video). I think this is the justification Sony is articulating to illustrate that 12MP > 61MP for low light. I think you are misinterpreting their message and generalizing it.
@@iDocSonic I have not tried this HDMI out method, but I think it is a great suggestion! Yeah unfortunately we are still at a point where the processor remains a big bottleneck in capturing high quality video.
Tony, great information when considering the ability to take and produce still images. However, in my experience the Sony A7Siii will out perform a larger megapixel camera like the 24MP Nikon Z6 when it comes to video in extreme low-light situations. I live in Alaska and use the A7Siii for Northern Lights video. I purchased the Nikon Z6 thinking that it would be able to perform at the same level as the Sony A7Sii/iii. They weren't even close. In this case the Sony A7Sii/iii out-performed the Nikon Z6!
Edward, you should try Canon ME20F-SH. It is said to be still better than a7s III. By the way, which camera - a7s II or a7s III - is better for low-light video?
The A7S 1 and 3 have better dynamic range performance at high ISOs than the A7R IV. Tony is right about high megapixel cameras producing better detail at high ISOs - but the test here needs to compare dynamic range performance and colour sensitivity as well if the purpose is to make conclusions about overall image quality.
Hey Tony, thanks for the content. It gets us thinking and reconsidering which is good. I do disagree though. Completely actually. To put it simply, there is a reason you use the A7Siii to film your videos. You did brush it off like it wasn't that big of a deal but let's think about it for a second. 1. A7Siii runs cooler while having a faster sensor readout providing far less rolling shutter as well as longer battery life all while giving a superior image in terms of color reproduction and micro contrast. 2. You can't zoom the grain on the 12MP and compare to a 60MP 1:1 and say the 12MP looks worse. You simply enlarged the grain of the 12MP, why? When viewed as a whole image side by side you can very clearly see the A7Siii has a much better image. 3. This same result can be seen with the Canon and Fuji samples as well. Better color and contrast with lower MP. 4. The A1 example was again, showing an 8k image vs 4k. The 8k will have more detail of course, but when viewing the images as a whole the A7Siii looks just as good or better and costs $3000 less and has all the advantages mentioned in #1 above. Also, the lower MP cameras use less storage and processing power and therefore require much less computer to process. All around a better experience. The 'trap' is more MP. 🤙
Excellent answer. Obvious "photographer bias" in the video. I remember when Philip Bloom tested the video coming out of the A6000 when it was released, and got the same results as the Canon C1, the outrage of the people who took a mortgage on the expensive camera in the comments. LOL
Although your result are clear, and I must agree, there is this but: every pixel on the image sensor is like a square bucket, and these have walls, walls that have a minimum thickness. These walls take away some light. The more pixels on one area, the more light is lost. Then again I don't know the numbers of how thick these walls and if they found a way to lose less light there. Do you know?
With the R5 mark II they figured out a way to increase the size of the pixel lenses (each pixel has its own lens on every sensor) so the lenses touch each other and there's no gap between them. Other than that. there's generally no published available data on the gaps between the pixel lenses.
One thing is missing. If you want to take bright details simultaneously to dark shadow details, lower megapixel sensor probably would have different exposure settings but also better detailed image. You may try to compare those cameras in video using SLog or CLog and zebras. You don't need a high ISO for that, you probably should use a lowest native one and compare it
Small pixels will always have greater noise PER PIXEL, BUT: Downsampling a higher resolution to a lower one on the resulting image will remove much of the noise due to the averaging lowpass effect, which is a big advantage of high res sensors. Anyway, high MPIX will only cause more processing work and more disk space requirement. In case you do not need the high resolution or your workflow involves downsampling anyway, there is no benefit from high MPIX sensors.
Dpreview has a web page where you can compare the same image taken from several different cameras. The lower mpx image usually looks cleaner, but there is an option to down sample them all to the same resolution. At the same resolution the noise difference virtually goes away.
This comparison is ridiculously unfair. Scaling the 12mp image to that of a 60mp image would OBVIOUSLY put the 12mp image to a huge disadvantage! ISO performance is not really tested well here considering you are just facing a wall with 2 static light sources behind you. The A7siii should be set at the correct ISO levels and Shutter speed where that map portrait is detailed enough. Then match the A7riv exposure levels to the A7siii result. The A7riv should result in a much darker image to match the noise levels to that of the A7siii.
I agree. With my A7R III it doesnt do bad in low light. But Ive also noticed I get more noise at ISO 400 or 640 than I do at ISO 1600. Maybe next video can go into detail about why this is so for those that dont understand how my higher ISO 1600 produces cleaner low light photos than at ISO 400 or 640.
Exposing right in camera using a high iso will always produce better results than using a lower iso and increasing the exposure in post-production, regardless of sensor. Provided you don’t clip any highlights.
If you shoot a higher iso in camera as opposed to stretching it in post, you avoid adding digital noise to the image. Also the camera is duel band iso settings. The higher band is less noisy than the low band iso settings.
I have also noticed this issue on my Sony A7Riii. Not sure why and trying to understand how I can get less noisy images when using 400-800 iso. Sometimes I shoot in auto mode and it still picks 400 iso and the image is very noisy. I use the Tamron 28-200mm lens. Not sure if that has anything to do.
@@iShootWild it has to do with the processor I believe. The images start getting clean again around 1200 ISO. This is why I said he should mention this in the next video. There are a lot of photographers that aeent awear of this.
I watched your old video on the same topic and this for a number of times... Couldn't understandbit until I understood your section on read speed and binning. Thanks Tony.
As an old school photographer, remember Kodachrome 25 😂😂, and broadcast engineer, I’ve learned a lot from your videos. You nailed it on HD recording sensor challenges. Keep up the good work and get more “toys”.
In the mid 70s I worked in the National Geographic photo labs. Kodak would sent samples of each Kodachrome run for testing and NGS would then buy the entire run and store it in freezers. It was the norm for the photographers to sometimes shoot several rolls just for one shot. Kodak created a second Kodachrome processing lab in Gaitherburg, MD so NGS could split all the incoming film into two batches, one sent there and the other to Rochester so in the event of a processing mishap there would be a back-up. The first step for magazine production was to make 35mm Ektachome duplicates as working copies for the editors. The lens on that copy camera cost more than my house at the time (houses were much cheaper then). To make prints for display 4x5 to 8 x 10 internegative would be made from the Kodachromes in a normal enlarger. The enlarger used to make the large display prints ran on tracks on the floor with the paper held to a wall by vacuum. The display area where the prints were evaluated had both 5000°K and 3200°K lighting so prints could be color balanced to the lighting where they would be displayed. I
@@TeddyCavachon Wow that’s a great experience. Envy you. Shot many rolls of Kodachrome 25 & 64. Learned proper exposure the hard way. Remember slide projectors😂. I still have vivid memories of the camera club in the early 70s. Professional photographers would mercilessly “tear” my pictures apart for exposure and composition. Speaking of cameras, my first was a Mamiya 1000. No electronics except for an inaccurate light meter. Did I mention also an old Roliflex. Great memories.
Excellent video, as someone who shoots handheld in low light environments, exactly what I needed to confirm my own experience. One follow up question. Does the stacked sensor in the Sony A1 make any difference shooting low light against say the new a7Rv. Thanks.
when i rented a sony a9 years ago that thing was nuts, could shoot in pretty shitty light conditions but still somehow managed to recover shadows and some exp when editing. wild camera. Also borrowed a gfx 50r from a friend for a shoot and that thing also recovered A LOT in editing. but the sony still felt like it took better night shots.
wouldn't many smaller pixels have more edges, thus losing some light on the edges between pixels? like, your results speak for themselves, but I am honestly shocked by them. great work ♥
@@UXXV Nope, sensors since 2007 have gapless micro lenses to eliminate that. And it was exceedly small that loss in the first place. Today if you compare bodies at the same output size, you will find that bodies like the R6 and R5 perform the same noise wise. But you gain all the advantages that extra resolution gives you.
@@johnrus7661 it wasnt the gaps between I was referring to, it was the edges. I remember how Fuji designed their hex SuperCCDs to try and combat gaps 20+ years ago.
@@UXXV It is clear you didn't read anything I said and know shit about photography. Edges, gaps. It's the same shit. There is not 100% coverage. You are a class act fool.
This isn't new information. Digitalrev TV made a comparison between the D4 and the D800, and one of the tests was low light performance. They took the same shot with both cameras, settings were the same. They zoomed in 1:1 on both images and yes, the D4 looked cleaner. But then they resized the image taken with the D800 to match that of the D4 and they looked extremely similar in noise level. This was done about 10 years ago.
Thanks for the video. There's actually one pretty important component to the Sony test that you may have missed, which is that the a7S III's sensor is not actually 12MP. Surprisingly, the IMX510 found in the a7S III is a 48MP quad-bayer sensor with 2x2 pixel binning, so the pixel size is actually fairly similar to that of the a7R IV's 61MP IMX451 sensor, and especially close to that of the a1's 50MP IMX610 sensor. Binning doesn't get around all of the penalties that are associated with smaller pixels (increased readout noise, lower fill factor, etc.), so it's not very surprising to see that the two cameras have incredibly similar noise levels. As far as I know this isn't documented anywhere by Sony.
This is the 1st time i read this anywhere, i want to upgrade my a7iii. I do a lot of low light events & astro, i was leaning towards A1 but that high MP worries me on the low light and dont want to rent all these cameras
Didn't this used to be true before sensors had gapless pixels? To use your bucket metaphor, some water used to be lost hitting the edge of the buckets. I believe it used to be the same with pixels in sensors.
True, but it must be said that Canon still uses front-side illuminated sensors on most of their cameras including R5 and R6, and they still do fine at high ISO.
if you assume the exact same sensor technology, won't the higher megapixel camera have more "wasted space" between the pixels? In the bucket analogy, if we assume some rain will fall between the small space between the buckets (non light gathering space between the pixels), the single bucket will capture more rain then 9 buckets... 9 buckets will capture more rain than 16 buckets... so on. This is why BSI sensors are better... there is less wasted space on the light gathering surface.
It's theoretically possible! We don't have numbers on how much wasted space there is. But if it was significant these tests would have shown a difference.
But it doesn't take in consideration that when you're using a high megapixel sensor you need to pair it with a very sharp lens in order to take advantage of the higher megapixel, so it makes the whole combo quite expensive, 20 or 24 megapixel on a full frame sensor for me are still great
The advantage will be less with a softer lens. But the image will still always be better on the higher resolution camera if you use the same lens. But today even entry level lenses tends to razor sharp around f5.6-8.
@@80-80. if so why for instance fuji made a list with the lenses that work better with the new 40 megapixel sensor... For me it's because the higher megapixel sensor will emphasize the cons of a lens, ad CA and low sharpness, it will make it appear even worse on a high megapixel sensor
@redis8298 I think you're right. A higher pixel density just might expose CA and LoCA of a lens that's not APO. Improper focus and exposure as well as camera shake and subject motion all rob IQ.
@@redis8298 Fuji published the list primarily to sell their high-end fixed focal lengths. It's true that you really benefit from them because they are already extremely sharp in general (comparable to Canon-L, Sigma art or Sony GM) but in practice, when you look at the images in a normalised view, you won't see any disadvantage in the image with a 40MP sensor. Of course, if you look at a 26MP and 40MP image and both at 150%, the crop on the 40MP is much smaller compared to the total image area because you are closer to a possible resolution limit of a lens. I have professionally tested all kinds of Fuji lenses with the X-H2 and the X-T5, you don't really need the new lenses, but if you don't have one yet, it makes sense to go for the latest technology.
There's so much more to it than pixel size and density! Each camera model is slightly different, even with the same sensor (the A7Rii, A7Riii and A99ii all share the same sensor, but don't have identical performance in low light), so a blanket statement like 'lower mega pixels is better in low light' is so banal it could only be terrible advice. I know if I compare my 2016 42mp A99ii to my 2006 10mp Sony A100 (CCD sensor!), the lower mega pixel camera is definitely not better in low light - ISO 400 in the old A100 is about equivalent to ISO 25600! Excellent work Tony, thanks!
But this largely ignores cost. Of course a higher mp sensor can have better low-light performance if more money is spent on it. If you gave two groups of engineers the same budget and goal of minimizing noise, and one had to make a 100mp full frame sensor and the other a 24mp full frame sensor and both had to be feasible to sell to consumers at the same price point, the 24mp sensor would absolutely have less noise.
Yes but the high mp cameras are usually more expensive e.g. R5 vs R6, the R3 has great low light performance because it's a more expensive pro camera with a high quality stacked sensor
I'm wondering, if you compare the 12MP to 60MP, and increase the size of the 12MP to match the size of 60MP, what would the noise quality be? I'm betting that the 60MP camera would have less noise than a 12MP image increase to 60MP. Which then mean that advertisements for 60MP camera are not wrong, as it does create less noise, but compare the noise of 12MP image vs 60MP image is comparing apples and oranges. I don't know if you have tested that, but if I had the equipment, that would be something I would check before stating that the noise reduction of a higher end camera is bunk. However, what I learn from this is that using a lower MP camera can be ideal for low light situations as it makes it easier for post-production.
Hi Tony, great video. I also find that the same amount of noise on two sensors for the same framing, will look less "noisy" with double the MPX as the noise is "half the size" for the high mpx camera in the final image even if all post processing is the same. I wonder if that made sense?
It does, noise in a high MPX sensor has "higher spatial frequency" so even if there is "more noise" it is occurring at a spatial frequency higher than the equivalent lower resolution sensor. Put another way, a low resolution sensor is nearly identical to a high resolution sensor where pixels are forcibly binned to the lower resolution. Only little detail left out is that there can be slight overhead in the sensor surface area, biasing the argument against what Tony is saying here, (by the bucket analogy, the bucket walls have some thickness that has to be considered, so a single large bucket collects a little more water... However, the spatial advantage allows noise reduction to overwhelmingly perform better with higher resolution anyways so Tony is correct for all practical purposes.
I come at this from an astrophotography angle. For a given technology level larger pixels (fewer pixels in a given area) gives you a better SNR. But if you use different tech (including noise reduction firmware, more modern pixel tech, etc.) then the smaller pixels (more pixels in the same area) may do as well or better than the larger pixels. So you cannot simply say that more pixels gives you less noise than do fewer pixels - or that fewer pixels means less noise. And when it comes to low light the issue isn't exactly noise which is paramount, it is the SNR. One other note? There are some cameras which use noise reduction circuitry. That makes normal testing look better but can end up tossing out some stars and similar which it thinks are noise but are actually signal. When that happens then the SNR for that pixel (or set of pixels) goes to zero and if your testing is simplistic it might look like the camera isn't noisy but it is doing at the expense of signal. It really isn't a simple issue and you can't accurately claim that in low light that more pixels is better - or worse. There really is much more that goes into it.
The example used is a highly detailed image with a lot of soft light bouncing off all the walls in the room, it showed the ability of the cameras to resolve text in a sorta low light setting. This wasn't a good test of how they perform when trying to capture things that often need high iso, street photography at night, wildlife photography under a thick canopy, photos with very dim lights and a strong color cast (candles, a campfire from 20ft away), literally anything where the ability to produce pleasant colors at high iso's might actually be tested. Sharpness and resolution aren't the be all end all of a photo, but this "test" made it seem like that's the case. Your bucket analogy doesn't consider things like filters being placed over each bucket which reduce the amount of rain they get at the edges, individual raindrops can't tell the bucket to send a signal to the processor so your chances of catching enough raindrops to send that signal goes up with the size of the bucket, the buckets aren't all the same size so they'll end up with more of one type of rain than the others (exacerbated by the last fact mentioned), the buckets would be spread out with gaps between them where rain will miss, smaller more densely packed buckets are harder to keep cool which sucks for extended use (and hotter buckets send more air to the processor), probably some others I can't think of. While buckets do have a pretty similar rain/air ratio nowadays, the ratio is an equation directly tied to the bucket pitch, meaning bigger buckets objectively give a better rain/air ratio. If you're gonna put TRUTH in the title, at least acknowledge the truth.
So were these images shot with the same exact lens, same exact fstop ? At the end of video there was some hand waving about “equivalent fstop”. No link in description. Love the energy Tony.
Tony, I know ISO is just q number as you said, but what does the uniquely high ISO of the A7S3 provide over the A1 provide in terms of image quality? Thanks in advance.
When you scaled the 12MP image did you scale it to the same size as the 60MP image or vice versa? If it's the former, there's a fatal flaw that you're upscaling a 12MP image to a 60MP image and any small noise will be magnified largely. If done vice versa, you're downscaling a 60MP image to a 12MP image which then makes it even more useless since then you're using a 60MP camera as a 12MP camera so what's the comparison being done for in the first place unless done pixel to pixel equally unless you plan to always print a 60MP image on a
1:23 its right but a picture taken at 100 iso and taken with a wrong shutter speed (so you have to push the iso in software) there you could see how the image really looks (before it was pushed) the problem here is that field that is bigger catch still detail here (but they can overlap or interfere) that problem is less noticeable when less light is coming also the 60 megapixel sensor can distinguish where the photon has "rained" on the small megapixels might got the same size and catch it but just maybe on the same sensel (and cant distinquish this time) thats hard to predict but more sensels or megapixels still have the advantage into both directions so what we can see is a yes and, no, "part in between" and hmm yes so i rather would choose the right physics smartphones have 1200 nanometers sensels cameras maybe less if the sensor is big 2400 nanometers but here is also the problem going below 1200 nanometers will be hard the lens would have a lot to do and the limit is almost reached the size for light is just 300-900 nanometers i would say we rather choose just the better physics the limit is almost reached with 1200 nanometers software and compression can play a role too
You are a major part of camera evolution and you have taught me a lot in the past. I've always liked your practical approach which is indisputable. Thanks
Including some BS about the multiplication of the f stop by the crop factor. A 50mm f2 apsc lens will have the same light transmission and bokeh as the 50mm full frame. The only difference being the apsc image will be smaller/cropped by a 1.5/1.6 factor. Everything else is the same, including the light transmission and bokeh quality. An f2 apsc is not an f3, because they let the same amount of light in.
correct me if I missed something., You didn't go out and shoot video in very lowlight conditions? I've seen footage from the A7s shot on a clear starlit night that was usable. Can you do that with the A7R IV?
Thanks for the video. Have you tried comparing high resolution cameras/sensors vs their low resolution counterparts for something like astrophotography and would (did) you get similar results?
I won’t believe this until I see a comparison with LESS LIGHT! And Tony you should know, that it’ll be much more interesting when the noise ruins the few photons collected by a small “bucket” while the more photons collected with the bigger “buckets” will have a far better noise to signal ratio thus showing more information. Who are you kidding,Tony?
Thanks for this! I've virtually given up on trying to correct the plethora of people that continue to parrot the erroneous concept of high mp being noisier for still images, even though DXOMARK numbers clearly demonstrate exactly what you're saying. When I got my a7Riv 2.5 yrs ago, and zoomed in to 100% pixels, I was initially shocked at the illusion of more noise, but quickly realized that the pixel level view was GREATLY magnifying any noise (compared to lower MP sensors) by taking a MUCH deeper dive into the image. This is a concept that has been so lost, to so many, that it's difficult to get folks to understand. Thanks again!
Tony confuses the issue here by oversimplifying. Sensors have a lot of factors to deal with besides light gathering. a 20 MP sensor will have different signal to noise depending on whether it is medium format, full frame or APSC. The medium format will have larger photo sites which gather more light for each pixel than the full frame, and likewise for full frame compared to APSC. More light per pixel equals better signal to noise ratio. Using his bucket example how much water was in ONE of the nine buckets compared to the one big bucket? The total light gathered by the entire sensor is irrelevant to the inherent noise. The noise is always there. In broad daylight the signal is strong enough to drown out any noise. It's only when we run out of leeway on the other ends of the triangle and must crank the exposure (gain) that the signal to noise becomes a force to contend with. As he mentions himself, but breezes right past. You have readout to contend with. Higher density sensors take more voltage and generate more heat than lower density ones. What is really going on is that the noise is finer on higher resolution sensors, because the sensor is finer. The manifestation of thermal randomness will have bigger chunks on a lower res sensor. For this to be scientific he would have explored ISO settings with compensation to keep them all the same exposure.
I love your videos. Aside from how informative they are, I really enjoy the technical explanation. Is it just me, or is there a sneaky, cheeky reference to Super Troopers? "Enhance" 🤣😂
I’d suggest that one exception to this rule would be astrophotography. A lower megapixel sensor will allow longer shutter speeds without visible star trailing. This allows either lower iso or more light gathering at the same iso (longer shutter), both of which are likely to produce a superior image. I know that’s a pretty niche scenario but a valid one nonetheless.
@@draughonc the important thing for me is ‘noticeable’ star trailing. A star trail will usually be caused because a star moves across your image, but stays the same height. So in a lower megapixel camera, a star might move 1pixel across, in the same amount of time that would be 5 pixels across in a sensor with 5 times resolution (an extreme example). A 1x2 trail will be far less noticeable than a 1x5 trail. Admittedly, the lower resolution will have other drawbacks. I’m just talking here about getting a cleaner image due to lower resolution, not what is overall a better way to capture an image.
Larger pixels have less detail, so you can leave the shutter open longer before seeing motion blur. In the case for stars the Asiii with the 55mm lens, you can use 8second shutter speed before seeing star trails. The A7riv would only be 6 seconds.
I think the main argument of high megapixels fans is that you can always scale the image down. Then star trails will not be more visible than with a low megapixel camera. But why buy a high megapixel camera and then scale it down? A high resolution restricts your photographic freedom a lot unless you want to scale the photos down in the end anyway. That really seems like a typical Tony argument, who also thinks that iPhones can compete with real cameras.
Tony thank you for your truth about these camera. Also the photo at 9:31 shows huge quality difference in the Fuil-frame vs APS-C photo. Also your wife is eternally beautiful and amazing. congrats on the two of you both being great in photography and also making most of your videos together, where I can see and feel the love and respect that you have for each other. Best wishes to both of you in ever way. I'd had 2 marriages and for years they were heaven on Earth, even though they ended at my decision, I would do them again they were that awesome. I am still friends with them both. I'm single now at my preference for now. Peace out. Love live amazing cameras and photography and women. LOL!
What about shutter speed and megapixels ? isn't it true that you can have longer shutter speeds with lower megapixels ? so you can freeze action the same on a 12mp sensor vs 60mp when having longer shutter speed.
This is the type of video that gave me a reason to subscribe. There may be other reasons to go with a lower-resolution sensor, especially for video, and that is for less rolling shutter or higher frame rates, depending on the specific read speeds of the sensors involved.
So for example, if you use a full frame 8K (assuming it’s 24mp) video then crop it to 4k in post, then take a full frame 4k (12mp) and don’t crop it, will they be equal? Obviously they will look different because of the different lens you use to capture the same frame but in terms of noise and light?
I always thought lower megapixels = better low light performance. The Nikon Z APS-C are very, very good at low light and that is a 21 Mp sensor I think.
The amount of light hitting the sensor is controled by aperture and shutter speed. So with the same amount of light collected the higher MP sensor does better. The lower MP pixels are bigger so they can collect more light per pixel than a higher MP sensor, before blowing out the highlights.
Thanks Tony, I did think larger pixels were more sensitive to light and had less noise but it makes sense that the most important thing to consider is total light across the sensor area. I guess one variable to consider is dual gain circuits. E.g. my black magic pocket 4k has dual native ISO at 400 and 3200 and it's a bit of a low light beast at 3200.
I have found on my higher megapixel camera that anything shot above iso 3200 has too much noise. It may have more detail, but I find the noise less desirable in the final product.
@@sexysilversurfer I have Lightroom, Topaz, and Luminar Neo. Sometimes Topaz will save a photo, but it depends. When I make timelapses from the photos I use Davinci Resolve to get rid of the noise and it works pretty good sometimes. I guess I just need a faster wider lens for astrophotography which is where I really want to go higher ISO but find the results disappointing.
Would be interested in your thoughts on high-ISO performance of the the Sony A7r V compared to it´s predecessor A7r IV. Available comparisons under identical condition (dpreview studio comparison tool) suggest that the A7r V is actually a step backwards, meaning visibly more noise at identical ISOs.
2:23 But to be fair, image on the left has way more consistent color across the whole area. While image on the right has tinted blobs all over the place.
I'm thinking of making the leap from my D3300 to a Z6... extreme use case could be taking kid pictures in ballet/theater performances... would the full frame sensor perform better in low light? Current lens NIKKOR AF-S 55-300 f/4.5-5.6G I know megapixels are not that different, but how about low light performance?
I went from d500 to z6 for similar reason (gymnastics). I think the overall IQ is slightly better with the z6, but that may be because I also got a better lens. BUT, the D500 tracked the focus way better AND it gave me 50% more reach (crop sensor). IMO, you will get better low light performance by getting a better lens (70-200 f2.8) on your D3300. Yes, my Z6 with the 70-200 is better, but if you upgrade the camera but still use an f5.6 lens, your images likely won't be any better than what you have.
Bucket analogy works against him. You would collect slightly more water with the one large bucket because, with the multiple buckets, there would be more loss when water hits the rims inbetween buckets (even with square buckets).
Correct. And the areas separating each pixel can take up a significant percentage of a sensors light-facing area. Backlit image sensors diminish this, but do not remove it entirely. So a sensor with fewer, large sensels will gather more light than a sensor with many, smaller sensels. In some cases, far more light. Tony knows this. Believe he only says these things to troll and drive engagement.
@@colinhoward2200 The analogy aspect is even generous because water that hits the rim of a bucket may partially splash in the bucket but light photons that hit the "rim" of a pixel are lost. So it's actually even worse than this analogy leads on.
@@miggetymike1 Lemme be pedantic to the max. Photons hitting the “rim” / edge / frame have a probabilistic chance of hitting the sensel - like a normal distribution curve. If pixel lens design, offset centers depending on axial distance to account for acute angle of incidence, and BSI are all at play then we can assume zero wall thickness is correct for the rain buckets. Or that the container walls are razor sharp and split water droplets probabilistically.
We’re talking apparent noise here (which is what matters anyway) because pixels produce noise. More of them=more noise. There’s read, bias, gain and thermal noise and its more of an issue in astrophotography and nightscape photography. Particularly in astrophotography with a telescope stacking hundreds of 5 min long exposures of very faint objects. In these situations you really see how noisy sensors can be. And high mp sensors have more of it, but they also produce more detailed images. At some point though the detailed resolution becomes useless because it can’t be perceived. It’s just noisier.
Always thought-provoking, Tony! Your conclusion comports with my own conclusion about low light performance owning first the a7s III, then the a7 IV, and now the a7R V.
Yes, I think that phenomenon is less a myth and more a tech limitation that has recently been ironed out. DSLRs suffered from this much more than the latest mirrorless offerings. For example, my 36mp Nikon D800 starts showing visible noise at around ISO 800, while my 16mp D4S never has noise at ISO 800, and only ever starts showing noise at around ISO 2500
There is also a possibility that at least in some instances in past people didn't scale images when comparing (they didn't upscale a smaller megapixel image to the size of a larger one), so they would view both low and high megapixel images at the pixel level, and when viewing them that way, of course higher megapixel camera will appear more noisy, since pixels are smaller, and thus the same number of pixels will cover smaller surface, and thus "gather" less light.
This realization hit me when I replaced my a7ii with an a7Rii. The “R” actually seemed to have better low light performance mainly because I could shoot at higher ISO’s and apply more aggressive noise reduction in post to compensate while losing very little detail in the bargain. (Being a BSI sensor probably didn’t hurt). 😉
I got the A7 II and I'm bitter about all the good stuff that the R II has going for it that I don't get, like the lower noise, being able to charge while running, and subject tracking that actually works (on the A7 II it completely drifts away immediately).
@@404Anymouse I have the 7ii and ater 7 years of owning it i can say that it is a nice camera to learn, but the autofocus sucks, working it as a videocamera is a nightmare, but well is the second A7line, almost 10 years have passed since its launch, we cannot ask more from it
In regards to video, wasn't the A7Sii considered the king of low-light? Trying to remember why that was the case if dual-native iso was a luxury released with the A7Siii.
For photographers that need to get photos out fast, such as when covering sporting or other live events, they won’t be processing RAW images. They will be using the image quality of the immediate JPEG file from the camera. In that scenario, don’t the lower megapixel sensors produce better low-light images compared to the high megapixel sensors? Maybe that is what the camera companies are using for the claimed low light advantage of lower megapixel cameras.
I've never found this to be the case. Especially now that high ISO cameras can pump out downsampled JPEG images, there are really no advantages that I've found to having fewer pixels. My 26 megapixel images from the A7R V are still cleaner than 24 native megapixels from A7C.
This is the exact conundrum I am currently mulling over, thanks for shedding some light on this. Interesting analogy with the buckets, but surly the issue is not the total amount of water (light) collected by the total combined area, but the amount of water collect by each bucket? And therefor the need for less gain where more is collected? Not to mention the higher percentage of “dead” space at the perimeter of each bucket. It’s a lot to get your head around and I’m looking at this from the perspective of a layman, so I guess the results speak for themselves. Is the improved detail of the higher megapixel sensor not simply the result of higher resolution?
Was your test image map offset printed? If so it was a pattern of CYMK dots and not an ideal subject for evaluating noise. A better test subject would be smooth white and gray 3D objects in lighting with a cross lighting in a large black draped space (to eliminate spill fill from lights) with starting with no fill to create a lighting ratio ratio / contrast range which exceeds the range of the sensor, exposing so the only the specular highlights are clipping (255) and the smooth white surfaces have eye dropper readings of 250. The white object will be rendered white-to-middle grey and the grey object will be reproduced from grey to the noise floor of the camera. The problem with your bucket analogy @ 4:40 is that digital cameras expose to keep the highlights from the highest area of reflectance in the scene below clipping. In high contrast scenes were luminance range is greater than sensor range the shutter will close and end the exposure before the “buckets” reproducing detailed shaded dark objects receive enough photos to record the detail so all that is amplified is the noise floor. Physically larger pixel sites (higher volume buckets) take longer to “fill to the brim” with photons in the highlights which give those in the shaded areas more time to gather enough photons to register an image, why sensors with larger photo sites in theory should have longer dynamic ranges. I say “in theory” because it is impossible to know how the signals are being processed. An empirical method I’ve used to test dynamic range and noise is to shoot a gray card in flat light filling the entire frame with a fast lens and start with lens wide open and a shutter speed which over-exposes the card to a point just under clipping (e.g. 250) you’d want to have a white non-specular highlight to have. Then bracket the exposures in one stop increments by first closing aperture and then doubling shutter speed if necessary until the spike the card creates on the histogram marches all the way from right to left. That test will measure the sensor range in f/stops and reveal via the spacing of the spikes how linear the sensor response is. The darker frames will reveal signal / noise level. Repeat test the for each ISO to evaluate how that variable affects exposure range, linearity of exposure and noise.
Thanks for the video and reminder that 'more is still more'. I had been considering an a7siii alongside my a7iv, but I'm minded to think that despite the higher cost that an a1 makes a better overall second body for a variety of reasons. Perhaps off topic but a pity that Sony's a1 flagship (which does an amazing amount) doesn't have a flip screen (trickier without a physical re-build - maybe in a1 mk ii) or firmware update features (it's prob just a firmware tweak needed) adding benefits like 'focus breathing'. Anyway for still and video it seems that an a1 is still streets ahead of an a7siii.
I have both Z6 and Z7 (first versions). The same story. Z6 is apparently better but if you resample images to the same size and apply noise reduction, Z7 smokes out Z6 big time. Although Z7 has more noise, the image retains more details and is generally better, especially colors in shadows that gets very weird toward magent-ish tone on Z6 at high ISOs, while stay very nice controlled and neutral on Z7.
I've noticed that myself in several cases (less MP is cleaner, but more MP is more detailed so the final result is similar), but the idea that "less MP is better for low light" still lived in my mind. Thanks Tony & Chelsea for this revelation!
Even when you throw science at some people, they will still stick to their guns and claim the lower resolution camera produces cleaner images. I've argued with people about this for years, but it's like arguing with a wall. They simply don't understand the scaling bit. They really can't understand that a 50mp camera sensor can always produce a 24mp image, while a 24mp sensor camera can never produce a 50mp image. It's totally lost on some people. The only true advantage of a smaller resolution sensor will be less work post processing to clean up high ISO images.
@@tubularificationed what is noiser, one huge basket collecting charge connected to one ADC or 4 smaller baskets connected to 4 ADC and averaged in software? Ofc it depend of ADC quality. If you get very good ADC it almost same, but in practice i very doubt. And ADC work in very hard conditions (crosstalk with adjucent lines/sensels, geometric tolerance of circuit elements must be 4 times finer, etc)
I'd say that 4 smaller baskets are privileged. - tolerance errors are evening out better, if you have 4 small buckets rather than 1 big one. - the same goes for dead pixels (or stuck pixels). Faults have only 25% effect rather than 100%. Repairing faults (e.g. interpolation between neighboring pixels) is also better, because at high resolutions, micro contrast edges are recorded more smoothly, which helps concealing mathematical interpolations. - all demosaicking / debayering artifacts also benefit from high resolutions, for related reasons. Because also demosaicking is some sort of interpolation between neighboring pixels. - 4 small pixels only accumulate 25% of electrical charges each, i.e. only would generate 25% of cross talk, and that would only affect 25% of the spatial neighborhood. But in general, ADCs and other electronics are so properly designed (already since long), that such issues play zero role since many years anyway. Which can be verified by inspecting any high-res photo taken during the last 10 years or so 😉 (or actually the entire era since when high-res sensors existed).
Really helpful video. A key point you mention is that some higher megapixel cameras skip pixels when filming because the processor isn't fast enough or cool enough to handle the light data. If we compare the FX3 (10.2 megapixels (effective) for movies) to the FX30 (20.1 megapixels (effective) for movies) do they use different processors ? Does the FX30 skip lines ? How does this relate to down sampling from 6 to 4 without pixel binning ?
I was just thinking should I sell my A7R4 for A74 due to the “poor performance” on low light on the A7R4, and this video just pops up on the recommend list on RUclips. Which helps a lot on my decision. Now I have decided to keep my A7R4(which is good because the 2nd hand price of A74 is higher than A7R4 now), just need find another way to light up the image..
I saw the method he used to get everyone confused, he scaled the high megapixel camera to 100% while scaling the lower megapixel camera to 400% but he never scaled the high megapixel camera to 400%
No he did scale it properly. The 12mp, image needs to be blown up more, because the computer thinks the photo is smaller then the 60mp photo. When in reality theyre both the same physical size.
I can't believe that in 2023 people still get confused by DPI/PPI and need absolute basics explained. Though actually it makes sense... No one ever prints anymore, and those fancy phones of whatever resolution screens and cameras do everything to obscure that knowledge for marketing reasons.
Detail and noise are 2 different things, we don’t want to see all the pores and bumps in the skin on a portrait, depends on how big you want to print your image. Higher megapixel camera don’t have less noise because they have more megapixel, it’s because the signal to noise ratio is better due to the technological improvement in the firmware. However, they still are noisy at low light conditions when a single photo site did not receive enough light or low signal value, under same low light conditions, the lower megapixel cameras collect more light per photo site or better signal value relatively, hence cleaner pixel. Even though 1 pixel in the low mp camera and 9 pixel in the high mp camera collect same light, for same total area, most of the pixel end up as noise in the high megapixel camera, whereas a quality pixel recorded in low mp camera.
Great, as always. Thanks. A few off topic comments perhaps for another episode: - Sharpness may not be everything. So here is another myth: 'Fewer, larger pixels are far better for dynamic range than more, smaller pixels'. - Moreover, for adapted lenses or more 'budget friendly' lenses (non professional) what is really the tradeoff between high megapixels and lower pixel pich?
Bless you Tony! I went with a Nikon D800 for shooting live concerts. To the standard thinking of the time that is the opposite of what is recommended. The D800 is awesome in low light and allows for cropping a lot better than some other cameras.
I suspect there are additional factors to consider regarding some comparisons. (I am particularly thinking of Math Photographer’s videos on this subject). For instance, the Leica SL2 (47MP), SL2S (24MP), M10-R (40MP), M10 Monochrom (40MP) and M11 (60MP) have been compared in various combinations regarding low light performance. The SL2S is both marketed and regarded as better at low light performance than the SL2, but the SL2S also has a newer and different sensor (BSI). The M10-R and M10 Monochrom, released at the same time, have the same resolution but the M10-R has the Bayer filter and concomitant demosaicing engine. Of course, the Monochrom has no color noise, and also presumably gathers more light due to the lack of a color filter. The M11 has much higher resolution, but also several years’ newer technology, including physical changes to sensor architecture. None of this takes away from Tony’s premise regarding MP count and sensor area. I am just saying that not all comparisons are truly apples to apples (or Cosmic Crispy to Red Delicious).
Sorry Tony, I have to disagree here in regards to “real world” low light usage. I set up an ARs3 & an A7r4 next to each other with identical settings & lenses & shot a Milky Way Time lapse, & the A7s3’s images were much clearer & less noisy. I understand what you are saying in regards to scaling( 12mp v 61mp) but I’m just giving you a “real world” example. Lol keep up the great work. 👍
“Cleaner”, “less noisy”… but you forgot about the last mitigating factor: “more detailed”. Perhaps in your astro-photography scenario, with just point sources of light, the “more detailed” part is not as differentiated. Just a guess…. and something to consider. It’s an ‘unusual’ subject matter to be sure. Not like rendering cursive scripts from a map.
@@billr3053 when viewing as a 4k render, the extra detail of the r4 was unusable to where the noise couldn’t be removed to a satisfactory result, so for astro timelapses viewing in 4k, the s3 gave a cleaner result without detail loss.
@@nashhall694 Interesting. Astro photography is my weak point. I’ll take your word for it but as I said that “unusual” subject matter may emphasize other strengths of a larger sensel-laden surface. As opposed to a wall map whose EV has been cranked to 18% gray level unlike a black sky for example that is supposed to remain black. I just thought of that. :)
He doesn't define what he means by "BEST" and this is a good illustration. From his tests he obviously considers capturing finer details to be the issue at play, which.. like the sensor has 5x the resolution, of course it's going to capture finer details. But there are so many factors at play, and the RUclips incentive structure is at odds with the fact that there is no one definition of "best" when it comes to image quality. Is it detail? Dynamic range? Noise? Accurate color? If one's objective is low light and high ISO, besides astrophotography being the perfect medium to test this in, one might also conclude that you'd be more interested in comparing noise or dynamic range (specifically the shadows). This video just.. it doesn't make any sense to me. It's trying to make a simple generalization about a multifactorial system based on a single factor. It's just doesn't demonstrate anything useful. Unless you consider showing that a higher resolution camera captures finer details than a lower resolution camera to be useful in some way.
I’m curious if this holds in *very* low light - like Milky Way photography under dark rural skies. I suppose I could compare my R7 to my R6 in crop mode. (But not anytime soon. Been a pretty wet winter.)
I wonder how does something like the Hasselblad X1DII or X2D compare vs full frame? they have high megapixel counts but have larger less dense photosites on the sensors. does that make for a cleaner image?
See I would like to know too. Does the logic of a larger “medium” sensor area reduce noise when compared to full frame if the MP are the same? Or is the result the same because the number of pixel are the same? Faster lens reduce noise...do large sensors?
If you are going to fiddle with noise reduction on every photo, then sure- I agree with you here. I shoot a lot of low-light PJ work and the noise processing in-camera for JPGs, for whatever reason, seems to perform better for me in the R6 than it did in my R7. I know it's also a crop sensor, but ultimately, I switched to the R6 from the R7 because it was far better in low light in my experience and the R7 was too fast for my needs. I can shoot 12,800 ISO on the R6 without significant issues, the R7 made it hard to shoot anything over 3200, and anything shot any higher than that certainly needed a heavy post-processing hand, as even out of camera JPGs were almost unusable.
Awesome video, I was really tensed about my a7rv purchase. Can you please make one comparison for ar7v 26mp vs 60mp image qauliity/noise comparison? Also, comparison to full frame vs apc video comparison?
3:05 - unpopular opinion - if you were to simply ask me 'which photo do you prefer', I actually would say the R6. Based on what I notice first and foremost, and based on my own priorities, the colours are more saturated and accurate, the contrast is noticeably better, the grain looks consistent and natural, the image looks less processed, and just noticeably crisper edges. The R5 image on the right is more detailed - granted - but my eye is far quicker drawn to the smeary look of Lightroom's noise reduction and the lower contrast, more processed look.
I'm a happy owner of an R5 and an R6, but the reason I reach for the R6 more readily for low light work, family dinners etc, is that the files are simply easier to look good with minimal work. I think some of it comes down to how much I dislike the look of Lightroom's luminance noise reduction. I am a big fan of Topaz DeNoise - which I use for wildlife photography but only on special images I intend to print or publish somewhere larger than social media - however the workflow simply isn't fast or convenient enough for bulk editing hundreds of photos from a party or other low light event.
I feel like one of the bigger reasons why people complain about high megapixel noise is on high MP sensors people feel very inclined to crop into images when they normally would not, which magnifies a lot of noise.
THIS.
The only reason is just because people don’t have enough money to buy both testing
This is accurate at least for me.
I got an A7Rii for the low light performance and learned this lesson the hard way. You can crop in for days but you lose low light performance.
I hate cropping for this very reason. At the same time, noise at ISO1600 for my 16MP camera is nowhere near the noise for my 46MP one.
I was guilty of this when I first started out
There is ONE exception to this. The only reason why Sony can put the second "high" 0db preamp at 12,800 ISO on the S-III is because those photosites are a very large 8+ microns in size. You wont ever see secondary preamps programed THAT high on photosites that are 4, 3 or 2 microns. You will see dual native ISO preamps on those smaller photosites...but they will be set MUCH closer together. (Like 800 and 2500 ISO or even closer than that) This is the main ability of that S-III sensor with giant photosites.
lol someone that knows. u should be making the video
That’s not the entire story. ISO12,800 on my FX3 is clean, but must be overexposed 1 to 2 stops, meaning it’s an equivalent ISO3200 from a different camera. ISO 25,600 is clean, as is ISO 51,200 - this is where the true strength of the dual stage of the Sony “S” cameras and (SLOG3). My A7R5 and the A7IV takes better sharper/more detailed video in the day time (And for the A7IV, basically it’s equally strong at night, while my A7R5 only performs as well in 4K30 APS-C Oversampled) than my FX3, especially if I shot in SLOG3 at the equivalent ISOs (Native 2500/800 in my A7R5). When I’m going out in very low conditions like a night market, concert etc where I would want ISO12,800 to ISO51,200, then that’s where I will definitely be taking my FX3 along.
The A7S III uses a nominal 48MP sensor binned down to 12MP, which unfortunately breaks your explanation. It is speculated that the A7S III is set up like this because each sub unit in the 4x4 "pixel" is tuned to a different sensitivity which, with some extra mojo, allows them to produce more dynamic range at the cost of some noise. The A7S II does use large pixels however. You can find this information online, the sensor was x-rayed (which is how this quirk was discovered) and it made the rounds a little while back.
So we’re the pictures taken in low light or was a high ISO used?
@@perilthecat It's more complex than that. If it's actually 48MP, then how is the A7S III readout speed so much faster than any comparable 48MP camera? This is evidenced by the A7S IIIs excellent rolling shutter.
Pixel binning could allow it, but then the noise performance would be terrible. Yet the noise on it, and the FX6 that shares the same sensor is extremely good.
Agree that Sony is using a novel technology, but the quantifiable evidence of the camera's output disproves the theory that it's binning, or has a traditional 48MP sensor.
Perhaps a fast, on-sensor subsample? Rapidly reading every pixel, discarding no light, and providing a lower resolution sample to the imaging pipeline? Whatever the answer, it's not a simple one.
I feel like most of this is about resolution only here. Not low light high iso noise. You can read Paris and Barcelona at 400% because the a7riv has 5 times the resolution of the a7siii. I'd like to see real pictures at a low light event. And see which are more usable than others.
True!
Those are real pictures aren't they? The point is that with real pictures, you are going to process them, and what you really care about is which gives a better image in the end, no? It's less useful to compare some isolated variable than it is to say 'which sensor provides better final images in low light'.
I agree. Low light capability is much more than end resolution using sliders - you have to consider the dynamic range and colour accuracy - both of wich are not possible to examine by photographing a world map.
Go to Dpreview website and use their Studio scene comparison tool found under Sample Images. There is a Size button, to zoom different megapixel pictures to the same visual size like Tony did.
You can compare upto 4 cameras at high iso say 25600, and the scene has gradient bars, detail text, and various elements to compare resolution, dynamic range, color response, etc…. In general Tony is correct.
Being a Canon user, I’ve held onto my 2016 50 MP 5DSR for that exact reason. Achievable images on high mpix are not worse, but often sharper with more detail. However, it also depends on how you Denoise. Some programs do a better job than others… so some may smear more detail on the noisier high megapix camera.
It is a valid point that the absolute difference between 50 mpix vs say 20 mpix is not leaps and bounds, seen only while pixel peeping. Printed or on a phone, will look similar.
Yep it’s always been the case and that’s why DXO low light (sports) score for high megapixels cameras is pretty much the same as the score for lower megapixel cameras, as DXO cleverly from day 1 id measuring normalized all sensors to 8MP. So 12mp sensor and 45mp sensor gets downscaled to same common number then measured.
The 12mp image should not be downscaled or upscaled just for comparisons. That's a completely unrealistic way to compare photos.
And that's why I look at the "screen" measurements and not just the "print" measurements.
Which is accurate for one type of delivery but not all use cases for cameras. Not everyone just shoots to delver 8MP printed photos. One of the main reasons why people use a higher resolution sensor is to crop so they don't need a APSC body or an optical extender which darkens the image by a stop. They like the flexibility to just shoot a photo and crop it as much as they want.
@@digitaldevigner4080 If you're "cropping as much as you want", and you're taking 24mp of the image, then the noise and light performance is the same as a crop sensor or a full frame with a teleconverter.
TCs and crop sensor bodies still seem to me to be the much cheaper option given that just 12MP makes for a very high quality image of a normal size, and that's about my only real gripe with the exorbitant price of high MP bodies. There are much cheaper ways to get the same result.
Some high res cams like a7riii, iv and v have better dxo sports rating than some 24mp models. Resolution alone might not account for noise by common sense.
Camera manufactures often don't disclose if the camera is line skipping. Or they line skip in higher demand shooting modes like 4k 60. In addition cameras also have dual gain and perform differently if you're filming at the native iso.
Tweener ISO's are just a push or pull from a native ISO, this applies to stills or vids. But the difference is slight. Bodies like the R5 don't skip lines at full 8K and can shoot RAW. So you get the same quality as in stills mode in vids.
No, you are wrong about this. This is not low light. I specialize in nightlife and live music, videography and photography. For a whole three to four year period I had to bring the a7R3 and the A7S2 with me every time I went out for a job. We're talking no light to some light in scenarios. Now I specifically own the A7S3 and A7IV to still accommodate those scenarios as I had to work a festival on New Year's Eve and then a warehouse after party. The higher megapixel cameras are better photo cameras all together, but they do not handle noise as well and attempting to recover detail or even denoising them will be comparably more difficult for instances where you really have limited amounts of light. They will simply introduce more noise that you cannot reduce as they attempt to find detail in a high ISO scenario. I have been doing this for almost 6 years at arguably the highest levels in this category of content creation. I have endless conversations with my peers comparing our loud outs. I can even tell you which Reds handle best in the same application.
I think many go wrong in the exact way that you show in the video: they zoom to 100% and see more noise. But if you apply denoising at full resolution and then export the image at the same resolution as a lower megapixel camera, you will get slightly better results.
The issue is that people like to crop their high mp images which magnifies the noise.
Also easy to fall into the trap of thinking a 400mm will become 600mp because of the high megapixels, but you're also magnifying the noise and the flaws in your lens
The other thing to consider is that with some of the new and amazing noise reduction software we no longer need to make such a trade off of removing noise and losing detail. I would much rather have the raw image with more detail and more noise to start my workflow.
So you want to go over layers with different AI noise reduction and have a complex workflow, extra software? I don't think so for me.
@@Hubieee price to pay for more detail
@@Hubieee try it once for raw denoisers and you would change your mind.
@Phillip Banes That's not true. Due to the nature of Bayer Pattern you lose a lot of potential when reducing the size early stages. Raw denoiser is always gonna be the best denoiser
@Phillip Banes no. You gotta try dxo raw denoiser. It’s free right now for 31 days, no watermarks.. It’s nothing like what we’re used to with denoise sliders in Lr/Ps/etc. Seriously game changing. I no longer feel the need to upgrade my camera body because of noise.
Lol, was it something I said?
This comparison is fine for the practical aspect of what it demonstrates, showing that higher resolution sensors capture higher resolution details, but that's as equally true of low ISO levels in bright light. (The A7S III has a 48MP sensor btw, it's binned down to 12MP)
Your bucket analogy is incomplete, because it fails to recognize the sources of noise and signal. The pixels of each sensor are not necessarily equal (ignoring size) when it comes to generating a signal from a certain number of photons. This is called quantum efficiency. To borrow a description of the benefits of a BSI vs FSI sensor that just happens to sum it up nicely, from a producer of astrophotography cameras (QHYCCD) where these nitpicks make a big difference:
"As a result, more incoming photons strike the photosensitive layer and more electrons are generated and captured in the pixel well. This ratio of photon to electron production is called quantum efficiency. The higher the quantum efficiency the more efficient the sensor is at converting photons to electrons and hence the more sensitive the sensor is to capturing an image of something dim."
If you're going to compare the ability to gather signal in low light between different sensors, you need to correct for the sensor's pixels' quantum efficiency. They don't all capture the same amount of rain even if they were the same size, to frame it in your analogy. And your analogy ignores that if each bucket needs 5 rain drops to generate 1 exposure value of signal, then the 9 small buckets will each generate less signal than the 1 large bucket, all other factors being equal. Which means even if you average those 9 buckets the total level of signal that they generate will be less than the 1 bucket spread over the same area.
As well, there is the reality of noise and its' sources. To again adapt your analogy, depending on the sensor you are talking about, the one bucket may or may not be introducing the same amount of noise as the nine. Each of the nine could be individually introducing more noise, and thus total noise, or less noise individually but more total noise, or the one could be introducing more noise total than the nine. Noise comes from a variety of areas, both analog and digital. It's a little more complicated to delve into things like input-referred read noise, read noise, analog and digital gain. There's the fact that each camera will have different numbers of ADCs (analog to digital converters) which kick in at different levels. That makes comparing between cameras at the same ISO not so much a test of whether big or small pixels are better for low light but whether that individual camera at that ISO is better than another individual camera at the same ISO.
There's also the fact that sensor density best matches itself with optical resolving capability at different focal lengths. Generally speaking, the more telephoto you go the lower the MP sensor you want, as you will otherwise be super-sampling to no benefit in resolution but with a noise penalty (this comes from astronomy where we're using 4000mm focal lengths, mind you). The opposite is true, denser sensors are generally a better match for wider angles.
Which is why this video, while demonstrating the slightly (im)practical results of it's simple comparisons, doesn't actually contend with what's going on. And I'm not exactly sure what myth is busted. Bigger pixels - assuming factors such as quantum efficiency are controlled for - are much more capable of generating better SNR. That's a completely different duck from whether high resolution sensors capture more resolution than lower resolution ones.
We may need better empirical analysis to understand more fully the relationship between pixel count and sensor noise for a given line of cameras from a given manufacturer. Tony's brief demonstration seems incomplete to me. It may be subject to experimenter bias and sampling error, as well as a flawed experimental design. We could start by comparing the images produced by the Sony a7R V (61 megapixels) vs the Sony a7 IV (33 megapixels) using a variety of lenses for the Sony E-mount, both entry level and state of the art. We could have professional photographers take hundreds of duplicate images in natural and artificial light, low light, indoors, and outside before examining the results, drawing a conclusion, and moving onto other camera comparisons from Cannon, Nikon, and Fujifilm. Taking this approach would be more scientific, in my opinion, than what Tony did for this video.
@@chesslover8829 I think it would be even more scientific to take cameras apart, scope the sensor analog output, the A/D data, truly measure noise at the analog and at the quantization stages, and do this for various light values. Aggregate results through various binning techniques (probabilistic energy distribution), and come up with envelope error bars that one must now, somehow, objectively, compare between cameras. In other words, go back to the design basics, being privy to the camera manufacturer’s secret pre-amp specifications S/N etc., manufacturing variances…. WOW… not easy.
@billr3053 you don’t need to do any of that actually, Bill Claff of photonstophotos has all the info you could ever want about any camera released in the last decade and a half. Tony has even used some of it when he dipped his toe in the ISO-invariant pool a few years back.
Now, just as then, he found a nugget of interest but made a video before looking far enough into it. He missed an opportunity to teach people about how to choose a sensor for their use case, instead choosing to make some faulty generalizations and then run some tests that confirm that higher res sensors produce higher resolution images. Truly groundbreaking
@chesslover8829 You don’t really even need to do any of that, all the relevant information exists already, and any side by side tests like Tony did will be inherently flawed. He’s trying to make a generalization about an oversimplification of how sensors compare. You can’t draw any consistent conclusions from that, because it ignores too many variables in the first place.
And he didn’t really have a cohesive “thing” he was trying to show. What does “best” mean? Lowest noise? Most detailed? Most dynamic range? Most sensitive in the shadows?
I’m honestly at a loss,
I’ll add an addendum, because it goes unremarked:
This “test” completely ignores the spatial filtering that manufacturers such as Sony bake into their RAWs, one purpose of which is to reduce noise in the final image. You can’t turn it off and it’s a different algorithm in every camera (and sometimes changes based on firmware version). To my knowledge the only manufacturer to forgo this was Canon in its DSLRs. This inherently makes a test such as this a fool’s errand, because there are too many uncontrolled variables.
And there’s no such thing as “equivalent aperture” for sensor noise and SNR, regardless of format. Imagine taking a photo with a D850, and then changing nothing else you drop it into DX mode and take another photo. What differences will you notice when you compare the two images at the same display resolution?
This video is misleading and a missed opportunity
I’m sorry Tony. The Sony A7S series has no equal for dim, exclusively naturally lit, and exclusively hand-held street photography. Just compare the published native ISOs of the various higher MP cameras to the A7 in the comparison. The A7S SEES IN THE DARK… resolving detail not visible to the unaided human eye. I was blown away by my first A7S as it demonstrated better night vision than a Litton image intensifier night scope. If you have any interest in night, hand-held street photography, rent any A7S for yourself and SEE (no pun intended). My Z9 rests comfortably at home when I go night hunting. There is no contest. A7S paired with a superfast wide angle prime (Sigma 20mm f1.4 Art), smiles all night…
Hi Tony.. I have an A7rii and an A7sii. I predominantly shoot astrophotography, and under real low light conditions, the A7s out performs the A7r every time. It's fair to say I shoot entirely in manual and have no requirement to crop or need for focus points, etc.
I tend to shoot single shot nightscape exposures, and the A7s will produce far less noise and, as a result, much cleaner images.
While I appreciate your points, I beg to differ after years of using both cameras in real-world low light conditions.
I think it is fair to say that for most cameras the video quality still remains better on lower megapixel sensor variants -- not because of the sensor altered features, but because of the limited readout ability with more megapixels!
So if you took HDMI out raw and processed externally where there were less limitations would it be better or equivalent? Or if processors get better and 48 megapixel would give you 8k video, it would be the best? Seems like it is all a processing problem and video has to process, stills are just capturing one image.
I do cover this but you can't really just say "most cameras" because it varies by specific models and the resolution/frame rate you film at.
@@TonyAndChelsea I am not discrediting you in this regard. You did acknowledge the readout limitations has an impact.
And yes, ultimately the comparison has to be apples to apples (ex: A7SIII is far superior vs A7RIV for 4K video). I think this is the justification Sony is articulating to illustrate that 12MP > 61MP for low light. I think you are misinterpreting their message and generalizing it.
@@iDocSonic I have not tried this HDMI out method, but I think it is a great suggestion!
Yeah unfortunately we are still at a point where the processor remains a big bottleneck in capturing high quality video.
@@iDocSonic no. The HDMI is output is still pixel skipped. You're not outputting 60 megapixels over HDMI. You're mixing up things.
Tony, great information when considering the ability to take and produce still images. However, in my experience the Sony A7Siii will out perform a larger megapixel camera like the 24MP Nikon Z6 when it comes to video in extreme low-light situations. I live in Alaska and use the A7Siii for Northern Lights video. I purchased the Nikon Z6 thinking that it would be able to perform at the same level as the Sony A7Sii/iii. They weren't even close. In this case the Sony A7Sii/iii out-performed the Nikon Z6!
Edward, you should try Canon ME20F-SH. It is said to be still better than a7s III. By the way, which camera - a7s II or a7s III - is better for low-light video?
The A7S 1 and 3 have better dynamic range performance at high ISOs than the A7R IV. Tony is right about high megapixel cameras producing better detail at high ISOs - but the test here needs to compare dynamic range performance and colour sensitivity as well if the purpose is to make conclusions about overall image quality.
Good point. Tony?
His test was composed of only reading text.
Hey Tony, thanks for the content. It gets us thinking and reconsidering which is good. I do disagree though. Completely actually. To put it simply, there is a reason you use the A7Siii to film your videos. You did brush it off like it wasn't that big of a deal but let's think about it for a second.
1. A7Siii runs cooler while having a faster sensor readout providing far less rolling shutter as well as longer battery life all while giving a superior image in terms of color reproduction and micro contrast.
2. You can't zoom the grain on the 12MP and compare to a 60MP 1:1 and say the 12MP looks worse. You simply enlarged the grain of the 12MP, why? When viewed as a whole image side by side you can very clearly see the A7Siii has a much better image.
3. This same result can be seen with the Canon and Fuji samples as well. Better color and contrast with lower MP.
4. The A1 example was again, showing an 8k image vs 4k. The 8k will have more detail of course, but when viewing the images as a whole the A7Siii looks just as good or better and costs $3000 less and has all the advantages mentioned in #1 above.
Also, the lower MP cameras use less storage and processing power and therefore require much less computer to process. All around a better experience. The 'trap' is more MP. 🤙
Excellent answer. Obvious "photographer bias" in the video. I remember when Philip Bloom tested the video coming out of the A6000 when it was released, and got the same results as the Canon C1, the outrage of the people who took a mortgage on the expensive camera in the comments. LOL
Although your result are clear, and I must agree, there is this but: every pixel on the image sensor is like a square bucket, and these have walls, walls that have a minimum thickness. These walls take away some light. The more pixels on one area, the more light is lost. Then again I don't know the numbers of how thick these walls and if they found a way to lose less light there. Do you know?
With the R5 mark II they figured out a way to increase the size of the pixel lenses (each pixel has its own lens on every sensor) so the lenses touch each other and there's no gap between them. Other than that. there's generally no published available data on the gaps between the pixel lenses.
One thing is missing. If you want to take bright details simultaneously to dark shadow details, lower megapixel sensor probably would have different exposure settings but also better detailed image. You may try to compare those cameras in video using SLog or CLog and zebras. You don't need a high ISO for that, you probably should use a lowest native one and compare it
What are you talking about, it's so disconnected I can't understand what you are trying to get at.
This literally sounds like it was written by an AI that was given a few camera terms to throw around.
Small pixels will always have greater noise PER PIXEL, BUT: Downsampling a higher resolution to a lower one on the resulting image will remove much of the noise due to the averaging lowpass effect, which is a big advantage of high res sensors. Anyway, high MPIX will only cause more processing work and more disk space requirement. In case you do not need the high resolution or your workflow involves downsampling anyway, there is no benefit from high MPIX sensors.
Dpreview has a web page where you can compare the same image taken from several different cameras. The lower mpx image usually looks cleaner, but there is an option to down sample them all to the same resolution. At the same resolution the noise difference virtually goes away.
You mean, comparing say a native 26mp apsc, and a 61 mp camera on apsc mode, they will have the same noise profile?
This comparison is ridiculously unfair. Scaling the 12mp image to that of a 60mp image would OBVIOUSLY put the 12mp image to a huge disadvantage!
ISO performance is not really tested well here considering you are just facing a wall with 2 static light sources behind you.
The A7siii should be set at the correct ISO levels and Shutter speed where that map portrait is detailed enough. Then match the A7riv exposure levels to the A7siii result. The A7riv should result in a much darker image to match the noise levels to that of the A7siii.
4:40 so the dividers between the pixels/buckets don't mean lost sensor surface?
I agree. With my A7R III it doesnt do bad in low light. But Ive also noticed I get more noise at ISO 400 or 640 than I do at ISO 1600. Maybe next video can go into detail about why this is so for those that dont understand how my higher ISO 1600 produces cleaner low light photos than at ISO 400 or 640.
Exposing right in camera using a high iso will always produce better results than using a lower iso and increasing the exposure in post-production, regardless of sensor. Provided you don’t clip any highlights.
@@EmilWall correft exposure or not my A7R III does better and less noise photos at 1600 and 3200 ISO than an 400, 640, or even 800 ISO.
If you shoot a higher iso in camera as opposed to stretching it in post, you avoid adding digital noise to the image. Also the camera is duel band iso settings. The higher band is less noisy than the low band iso settings.
I have also noticed this issue on my Sony A7Riii. Not sure why and trying to understand how I can get less noisy images when using 400-800 iso. Sometimes I shoot in auto mode and it still picks 400 iso and the image is very noisy. I use the Tamron 28-200mm lens. Not sure if that has anything to do.
@@iShootWild it has to do with the processor I believe. The images start getting clean again around 1200 ISO. This is why I said he should mention this in the next video. There are a lot of photographers that aeent awear of this.
I watched your old video on the same topic and this for a number of times... Couldn't understandbit until I understood your section on read speed and binning. Thanks Tony.
Agree and e.g. DXOmark has proven this in their test charts for years as well ... when you compare the print and not the screen measurement.
As an old school photographer, remember Kodachrome 25 😂😂, and broadcast engineer, I’ve learned a lot from your videos. You nailed it on HD recording sensor challenges. Keep up the good work and get more “toys”.
In the mid 70s I worked in the National Geographic photo labs. Kodak would sent samples of each Kodachrome run for testing and NGS would then buy the entire run and store it in freezers. It was the norm for the photographers to sometimes shoot several rolls just for one shot. Kodak created a second Kodachrome processing lab in Gaitherburg, MD so NGS could split all the incoming film into two batches, one sent there and the other to Rochester so in the event of a processing mishap there would be a back-up.
The first step for magazine production was to make 35mm Ektachome duplicates as working copies for the editors. The lens on that copy camera cost more than my house at the time (houses were much cheaper then). To make prints for display 4x5 to 8 x 10 internegative would be made from the Kodachromes in a normal enlarger. The enlarger used to make the large display prints ran on tracks on the floor with the paper held to a wall by vacuum. The display area where the prints were evaluated had both 5000°K and 3200°K lighting so prints could be color balanced to the lighting where they would be displayed.
I
@@TeddyCavachon Wow that’s a great experience. Envy you. Shot many rolls of Kodachrome 25 & 64. Learned proper exposure the hard way.
Remember slide projectors😂. I still have vivid memories of the camera club in the early 70s. Professional photographers would mercilessly “tear” my pictures apart for exposure and composition.
Speaking of cameras, my first was a Mamiya 1000. No electronics except for an inaccurate light meter. Did I mention also an old Roliflex. Great memories.
Excellent video, as someone who shoots handheld in low light environments, exactly what I needed to confirm my own experience. One follow up question. Does the stacked sensor in the Sony A1 make any difference shooting low light against say the new a7Rv. Thanks.
when i rented a sony a9 years ago that thing was nuts, could shoot in pretty shitty light conditions but still somehow managed to recover shadows and some exp when editing. wild camera. Also borrowed a gfx 50r from a friend for a shoot and that thing also recovered A LOT in editing. but the sony still felt like it took better night shots.
wouldn't many smaller pixels have more edges, thus losing some light on the edges between pixels? like, your results speak for themselves, but I am honestly shocked by them. great work ♥
Yes. 9 buckets would not collect the same amount of water due to the edges.
@@UXXV Except, that the added resolution probably more than makes up for the gaps!
@@UXXV Nope, sensors since 2007 have gapless micro lenses to eliminate that. And it was exceedly small that loss in the first place. Today if you compare bodies at the same output size, you will find that bodies like the R6 and R5 perform the same noise wise. But you gain all the advantages that extra resolution gives you.
@@johnrus7661 it wasnt the gaps between I was referring to, it was the edges. I remember how Fuji designed their hex SuperCCDs to try and combat gaps 20+ years ago.
@@UXXV It is clear you didn't read anything I said and know shit about photography.
Edges, gaps. It's the same shit. There is not 100% coverage.
You are a class act fool.
This isn't new information. Digitalrev TV made a comparison between the D4 and the D800, and one of the tests was low light performance. They took the same shot with both cameras, settings were the same. They zoomed in 1:1 on both images and yes, the D4 looked cleaner. But then they resized the image taken with the D800 to match that of the D4 and they looked extremely similar in noise level. This was done about 10 years ago.
I remember that video! Classic Digitalrev TV. I miss those days. Where's Alamby ... jk lol
That are different sensors
Thanks for the video. There's actually one pretty important component to the Sony test that you may have missed, which is that the a7S III's sensor is not actually 12MP. Surprisingly, the IMX510 found in the a7S III is a 48MP quad-bayer sensor with 2x2 pixel binning, so the pixel size is actually fairly similar to that of the a7R IV's 61MP IMX451 sensor, and especially close to that of the a1's 50MP IMX610 sensor. Binning doesn't get around all of the penalties that are associated with smaller pixels (increased readout noise, lower fill factor, etc.), so it's not very surprising to see that the two cameras have incredibly similar noise levels. As far as I know this isn't documented anywhere by Sony.
This is the 1st time i read this anywhere, i want to upgrade my a7iii. I do a lot of low light events & astro, i was leaning towards A1 but that high MP worries me on the low light and dont want to rent all these cameras
Didn't this used to be true before sensors had gapless pixels? To use your bucket metaphor, some water used to be lost hitting the edge of the buckets. I believe it used to be the same with pixels in sensors.
True, but it must be said that Canon still uses front-side illuminated sensors on most of their cameras including R5 and R6, and they still do fine at high ISO.
if you assume the exact same sensor technology, won't the higher megapixel camera have more "wasted space" between the pixels? In the bucket analogy, if we assume some rain will fall between the small space between the buckets (non light gathering space between the pixels), the single bucket will capture more rain then 9 buckets... 9 buckets will capture more rain than 16 buckets... so on. This is why BSI sensors are better... there is less wasted space on the light gathering surface.
It's theoretically possible! We don't have numbers on how much wasted space there is. But if it was significant these tests would have shown a difference.
But it doesn't take in consideration that when you're using a high megapixel sensor you need to pair it with a very sharp lens in order to take advantage of the higher megapixel, so it makes the whole combo quite expensive, 20 or 24 megapixel on a full frame sensor for me are still great
The advantage will be less with a softer lens. But the image will still always be better on the higher resolution camera if you use the same lens. But today even entry level lenses tends to razor sharp around f5.6-8.
@@80-80. if so why for instance fuji made a list with the lenses that work better with the new 40 megapixel sensor... For me it's because the higher megapixel sensor will emphasize the cons of a lens, ad CA and low sharpness, it will make it appear even worse on a high megapixel sensor
@redis8298 I think you're right. A higher pixel density just might expose CA and LoCA of a lens that's not APO. Improper focus and exposure as well as camera shake and subject motion all rob IQ.
@@redis8298 Fuji published the list primarily to sell their high-end fixed focal lengths. It's true that you really benefit from them because they are already extremely sharp in general (comparable to Canon-L, Sigma art or Sony GM) but in practice, when you look at the images in a normalised view, you won't see any disadvantage in the image with a 40MP sensor. Of course, if you look at a 26MP and 40MP image and both at 150%, the crop on the 40MP is much smaller compared to the total image area because you are closer to a possible resolution limit of a lens. I have professionally tested all kinds of Fuji lenses with the X-H2 and the X-T5, you don't really need the new lenses, but if you don't have one yet, it makes sense to go for the latest technology.
There's so much more to it than pixel size and density! Each camera model is slightly different, even with the same sensor (the A7Rii, A7Riii and A99ii all share the same sensor, but don't have identical performance in low light), so a blanket statement like 'lower mega pixels is better in low light' is so banal it could only be terrible advice. I know if I compare my 2016 42mp A99ii to my 2006 10mp Sony A100 (CCD sensor!), the lower mega pixel camera is definitely not better in low light - ISO 400 in the old A100 is about equivalent to ISO 25600! Excellent work Tony, thanks!
But this largely ignores cost. Of course a higher mp sensor can have better low-light performance if more money is spent on it. If you gave two groups of engineers the same budget and goal of minimizing noise, and one had to make a 100mp full frame sensor and the other a 24mp full frame sensor and both had to be feasible to sell to consumers at the same price point, the 24mp sensor would absolutely have less noise.
Yes but the high mp cameras are usually more expensive e.g. R5 vs R6, the R3 has great low light performance because it's a more expensive pro camera with a high quality stacked sensor
I'm wondering, if you compare the 12MP to 60MP, and increase the size of the 12MP to match the size of 60MP, what would the noise quality be? I'm betting that the 60MP camera would have less noise than a 12MP image increase to 60MP. Which then mean that advertisements for 60MP camera are not wrong, as it does create less noise, but compare the noise of 12MP image vs 60MP image is comparing apples and oranges. I don't know if you have tested that, but if I had the equipment, that would be something I would check before stating that the noise reduction of a higher end camera is bunk. However, what I learn from this is that using a lower MP camera can be ideal for low light situations as it makes it easier for post-production.
Hi Tony, great video. I also find that the same amount of noise on two sensors for the same framing, will look less "noisy" with double the MPX as the noise is "half the size" for the high mpx camera in the final image even if all post processing is the same. I wonder if that made sense?
It does, noise in a high MPX sensor has "higher spatial frequency" so even if there is "more noise" it is occurring at a spatial frequency higher than the equivalent lower resolution sensor. Put another way, a low resolution sensor is nearly identical to a high resolution sensor where pixels are forcibly binned to the lower resolution. Only little detail left out is that there can be slight overhead in the sensor surface area, biasing the argument against what Tony is saying here, (by the bucket analogy, the bucket walls have some thickness that has to be considered, so a single large bucket collects a little more water... However, the spatial advantage allows noise reduction to overwhelmingly perform better with higher resolution anyways so Tony is correct for all practical purposes.
I come at this from an astrophotography angle. For a given technology level larger pixels (fewer pixels in a given area) gives you a better SNR. But if you use different tech (including noise reduction firmware, more modern pixel tech, etc.) then the smaller pixels (more pixels in the same area) may do as well or better than the larger pixels.
So you cannot simply say that more pixels gives you less noise than do fewer pixels - or that fewer pixels means less noise.
And when it comes to low light the issue isn't exactly noise which is paramount, it is the SNR.
One other note? There are some cameras which use noise reduction circuitry. That makes normal testing look better but can end up tossing out some stars and similar which it thinks are noise but are actually signal. When that happens then the SNR for that pixel (or set of pixels) goes to zero and if your testing is simplistic it might look like the camera isn't noisy but it is doing at the expense of signal.
It really isn't a simple issue and you can't accurately claim that in low light that more pixels is better - or worse. There really is much more that goes into it.
The example used is a highly detailed image with a lot of soft light bouncing off all the walls in the room, it showed the ability of the cameras to resolve text in a sorta low light setting. This wasn't a good test of how they perform when trying to capture things that often need high iso, street photography at night, wildlife photography under a thick canopy, photos with very dim lights and a strong color cast (candles, a campfire from 20ft away), literally anything where the ability to produce pleasant colors at high iso's might actually be tested. Sharpness and resolution aren't the be all end all of a photo, but this "test" made it seem like that's the case.
Your bucket analogy doesn't consider things like filters being placed over each bucket which reduce the amount of rain they get at the edges, individual raindrops can't tell the bucket to send a signal to the processor so your chances of catching enough raindrops to send that signal goes up with the size of the bucket, the buckets aren't all the same size so they'll end up with more of one type of rain than the others (exacerbated by the last fact mentioned), the buckets would be spread out with gaps between them where rain will miss, smaller more densely packed buckets are harder to keep cool which sucks for extended use (and hotter buckets send more air to the processor), probably some others I can't think of. While buckets do have a pretty similar rain/air ratio nowadays, the ratio is an equation directly tied to the bucket pitch, meaning bigger buckets objectively give a better rain/air ratio.
If you're gonna put TRUTH in the title, at least acknowledge the truth.
So were these images shot with the same exact lens, same exact fstop ? At the end of video there was some hand waving about “equivalent fstop”. No link in description. Love the energy Tony.
On the Fujifilm GFX 100s (102mp) the noise is extremely small compared to the size of the photos. At ISO 12800 the image quality is insane.
well.. it's also a medium format camera.... completely different sensor size...
I'd actually expect the higher MP camera to show less noise after rescaling to the lower MP count, because noise is averaged out.
Tony, I know ISO is just q number as you said, but what does the uniquely high ISO of the A7S3 provide over the A1 provide in terms of image quality? Thanks in advance.
Could be higher dynamic range at higher ISO's...
Can you compare crop and full-frame sensor with same lens to find out field-of-view, Noise, light levels
When you scaled the 12MP image did you scale it to the same size as the 60MP image or vice versa? If it's the former, there's a fatal flaw that you're upscaling a 12MP image to a 60MP image and any small noise will be magnified largely.
If done vice versa, you're downscaling a 60MP image to a 12MP image which then makes it even more useless since then you're using a 60MP camera as a 12MP camera so what's the comparison being done for in the first place unless done pixel to pixel equally unless you plan to always print a 60MP image on a
lol what are you on about? its not 2001 anymore
@@heat1235 : What are YOU on about? You want higher MP, why not use the new 104 MP sensor on your smartphone instead? ;)
1:23 its right but a picture taken at 100 iso and taken with a wrong shutter speed (so you have to push the iso in software) there you could see how the image really looks (before it was pushed)
the problem here is that field that is bigger catch still detail here (but they can overlap or interfere)
that problem is less noticeable when less light is coming
also the 60 megapixel sensor can distinguish where the photon has "rained" on
the small megapixels might got the same size and catch it but just maybe on the same sensel (and cant distinquish this time) thats hard to predict
but more sensels or megapixels still have the advantage into both directions
so what we can see is a yes and, no, "part in between" and hmm yes
so i rather would choose the right physics smartphones have 1200 nanometers sensels cameras maybe less if the sensor is big 2400 nanometers
but here is also the problem going below 1200 nanometers will be hard the lens would have a lot to do
and the limit is almost reached the size for light is just 300-900 nanometers
i would say we rather choose just the better physics
the limit is almost reached with 1200 nanometers
software and compression can play a role too
You are a major part of camera evolution and you have taught me a lot in the past.
I've always liked your practical approach which is indisputable.
Thanks
Including some BS about the multiplication of the f stop by the crop factor. A 50mm f2 apsc lens will have the same light transmission and bokeh as the 50mm full frame. The only difference being the apsc image will be smaller/cropped by a 1.5/1.6 factor. Everything else is the same, including the light transmission and bokeh quality. An f2 apsc is not an f3, because they let the same amount of light in.
@@itacatv2146 bokeh is relative to depth of field though, and depth of field Is changed when going from APS C to FF...
Great news, as a wedding photographer I can buy the Z8 after all. 😉
@@acronymphotography Nobody cares but congrats I guess.
correct me if I missed something., You didn't go out and shoot video in very lowlight conditions? I've seen footage from the A7s shot on a clear starlit night that was usable. Can you do that with the A7R IV?
Thanks for the video. Have you tried comparing high resolution cameras/sensors vs their low resolution counterparts for something like astrophotography and would (did) you get similar results?
DSLR for astro always go for lower megapixel. You'll get lower read noise and better dynamic range.
@@MattDvc did you watch the video? 😂
@@EverythingCameFromNothing my comment is not based on the video, but experience using 24mp z6 and 45mp z7 for astrophotography.
@@MattDvc Did you compare the images at the same magnification OR equivalent sizes?
I’d be interested to see some of your examples
@@EverythingCameFromNothing it's not how astrophotography works mate.
Was the dynamic range setting the same on the compared cameras while taking the photos? Auto? 100%? 400%?
I won’t believe this until I see a comparison with LESS LIGHT! And Tony you should know, that it’ll be much more interesting when the noise ruins the few photons collected by a small “bucket” while the more photons collected with the bigger “buckets” will have a far better noise to signal ratio thus showing more information. Who are you kidding,Tony?
There is more noise with higher MP but after NR retains more DETAIL than low MP
Thanks for this! I've virtually given up on trying to correct the plethora of people that continue to parrot the erroneous concept of high mp being noisier for still images, even though DXOMARK numbers clearly demonstrate exactly what you're saying. When I got my a7Riv 2.5 yrs ago, and zoomed in to 100% pixels, I was initially shocked at the illusion of more noise, but quickly realized that the pixel level view was GREATLY magnifying any noise (compared to lower MP sensors) by taking a MUCH deeper dive into the image. This is a concept that has been so lost, to so many, that it's difficult to get folks to understand. Thanks again!
Tony confuses the issue here by oversimplifying.
Sensors have a lot of factors to deal with besides light gathering.
a 20 MP sensor will have different signal to noise depending on whether it is medium format, full frame or APSC.
The medium format will have larger photo sites which gather more light for each pixel than the full frame, and likewise for full frame compared to APSC.
More light per pixel equals better signal to noise ratio.
Using his bucket example how much water was in ONE of the nine buckets compared to the one big bucket?
The total light gathered by the entire sensor is irrelevant to the inherent noise.
The noise is always there. In broad daylight the signal is strong enough to drown out any noise.
It's only when we run out of leeway on the other ends of the triangle and must crank the exposure (gain) that the signal to noise becomes a force to contend with.
As he mentions himself, but breezes right past. You have readout to contend with.
Higher density sensors take more voltage and generate more heat than lower density ones.
What is really going on is that the noise is finer on higher resolution sensors, because the sensor is finer.
The manifestation of thermal randomness will have bigger chunks on a lower res sensor.
For this to be scientific he would have explored ISO settings with compensation to keep them all the same exposure.
I love your videos. Aside from how informative they are, I really enjoy the technical explanation. Is it just me, or is there a sneaky, cheeky reference to Super Troopers? "Enhance" 🤣😂
I’d suggest that one exception to this rule would be astrophotography. A lower megapixel sensor will allow longer shutter speeds without visible star trailing. This allows either lower iso or more light gathering at the same iso (longer shutter), both of which are likely to produce a superior image. I know that’s a pretty niche scenario but a valid one nonetheless.
How is the size of pixels related to shutter speed?
@@draughonc the important thing for me is ‘noticeable’ star trailing. A star trail will usually be caused because a star moves across your image, but stays the same height. So in a lower megapixel camera, a star might move 1pixel across, in the same amount of time that would be 5 pixels across in a sensor with 5 times resolution (an extreme example). A 1x2 trail will be far less noticeable than a 1x5 trail. Admittedly, the lower resolution will have other drawbacks. I’m just talking here about getting a cleaner image due to lower resolution, not what is overall a better way to capture an image.
Larger pixels have less detail, so you can leave the shutter open longer before seeing motion blur. In the case for stars the Asiii with the 55mm lens, you can use 8second shutter speed before seeing star trails. The A7riv would only be 6 seconds.
I think the main argument of high megapixels fans is that you can always scale the image down. Then star trails will not be more visible than with a low megapixel camera. But why buy a high megapixel camera and then scale it down? A high resolution restricts your photographic freedom a lot unless you want to scale the photos down in the end anyway.
That really seems like a typical Tony argument, who also thinks that iPhones can compete with real cameras.
Tony thank you for your truth about these camera. Also the photo at 9:31 shows huge quality difference in the Fuil-frame vs APS-C photo. Also your wife is eternally beautiful and amazing. congrats on the two of you both being great in photography and also making most of your videos together, where I can see and feel the love and respect that you have for each other. Best wishes to both of you in ever way. I'd had 2 marriages and for years they were heaven on Earth, even though they ended at my decision, I would do them again they were that awesome. I am still friends with them both. I'm single now at my preference for now. Peace out. Love live amazing cameras and photography and women. LOL!
PLEASE talk about how Dual Native ISO's compare with high & low MP Cameras in regard to noise.
What about shutter speed and megapixels ? isn't it true that you can have longer shutter speeds with lower megapixels ?
so you can freeze action the same on a 12mp sensor vs 60mp when having longer shutter speed.
The movement is more obvious on high megapixel cameras but you could always scale the images down and get the same effect.
This is the type of video that gave me a reason to subscribe. There may be other reasons to go with a lower-resolution sensor, especially for video, and that is for less rolling shutter or higher frame rates, depending on the specific read speeds of the sensors involved.
So for example, if you use a full frame 8K (assuming it’s 24mp) video then crop it to 4k in post, then take a full frame 4k (12mp) and don’t crop it, will they be equal? Obviously they will look different because of the different lens you use to capture the same frame but in terms of noise and light?
I always thought lower megapixels = better low light performance. The Nikon Z APS-C are very, very good at low light and that is a 21 Mp sensor I think.
The amount of light hitting the sensor is controled by aperture and shutter speed. So with the same amount of light collected the higher MP sensor does better. The lower MP pixels are bigger so they can collect more light per pixel than a higher MP sensor, before blowing out the highlights.
I decided the R3 wasn't because of NP,but generally for speed. Thanks for the advice and will use it in the future.
Thanks Tony, I did think larger pixels were more sensitive to light and had less noise but it makes sense that the most important thing to consider is total light across the sensor area. I guess one variable to consider is dual gain circuits. E.g. my black magic pocket 4k has dual native ISO at 400 and 3200 and it's a bit of a low light beast at 3200.
I have found on my higher megapixel camera that anything shot above iso 3200 has too much noise. It may have more detail, but I find the noise less desirable in the final product.
@@ElMundoDurowhat noise reduction software are you using?
@@sexysilversurfer I have Lightroom, Topaz, and Luminar Neo. Sometimes Topaz will save a photo, but it depends. When I make timelapses from the photos I use Davinci Resolve to get rid of the noise and it works pretty good sometimes. I guess I just need a faster wider lens for astrophotography which is where I really want to go higher ISO but find the results disappointing.
@@ElMundoDuro looks like you have tried the main options for noise reduction.
@@sexysilversurfer except the best one… DxO Pureraw. Doesn’t leave images looking as plasticy as topaz, and seems to keep more detail.
Would be interested in your thoughts on high-ISO performance of the the Sony A7r V compared to it´s predecessor A7r IV. Available comparisons under identical condition (dpreview studio comparison tool) suggest that the A7r V is actually a step backwards, meaning visibly more noise at identical ISOs.
2:23 But to be fair, image on the left has way more consistent color across the whole area. While image on the right has tinted blobs all over the place.
A brilliant, yet simple analogy to explain light collection via pixels! Made things clearer for me at any rate. Great work.
I'm thinking of making the leap from my D3300 to a Z6... extreme use case could be taking kid pictures in ballet/theater performances... would the full frame sensor perform better in low light? Current lens
NIKKOR AF-S 55-300 f/4.5-5.6G
I know megapixels are not that different, but how about low light performance?
I went from d500 to z6 for similar reason (gymnastics). I think the overall IQ is slightly better with the z6, but that may be because I also got a better lens. BUT, the D500 tracked the focus way better AND it gave me 50% more reach (crop sensor). IMO, you will get better low light performance by getting a better lens (70-200 f2.8) on your D3300. Yes, my Z6 with the 70-200 is better, but if you upgrade the camera but still use an f5.6 lens, your images likely won't be any better than what you have.
Bucket analogy works against him. You would collect slightly more water with the one large bucket because, with the multiple buckets, there would be more loss when water hits the rims inbetween buckets (even with square buckets).
You're right.
It was only an analogy - you have tried to over-think it.
Correct. And the areas separating each pixel can take up a significant percentage of a sensors light-facing area. Backlit image sensors diminish this, but do not remove it entirely. So a sensor with fewer, large sensels will gather more light than a sensor with many, smaller sensels. In some cases, far more light.
Tony knows this. Believe he only says these things to troll and drive engagement.
@@colinhoward2200 The analogy aspect is even generous because water that hits the rim of a bucket may partially splash in the bucket but light photons that hit the "rim" of a pixel are lost. So it's actually even worse than this analogy leads on.
@@miggetymike1 Lemme be pedantic to the max. Photons hitting the “rim” / edge / frame have a probabilistic chance of hitting the sensel - like a normal distribution curve. If pixel lens design, offset centers depending on axial distance to account for acute angle of incidence, and BSI are all at play then we can assume zero wall thickness is correct for the rain buckets. Or that the container walls are razor sharp and split water droplets probabilistically.
We’re talking apparent noise here (which is what matters anyway) because pixels produce noise. More of them=more noise. There’s read, bias, gain and thermal noise and its more of an issue in astrophotography and nightscape photography. Particularly in astrophotography with a telescope stacking hundreds of 5 min long exposures of very faint objects. In these situations you really see how noisy sensors can be. And high mp sensors have more of it, but they also produce more detailed images. At some point though the detailed resolution becomes useless because it can’t be perceived. It’s just noisier.
Always thought-provoking, Tony! Your conclusion comports with my own conclusion about low light performance owning first the a7s III, then the a7 IV, and now the a7R V.
Well if two of the biggest names on youtube regarding photography both agree on this then that is good enough for me!!!
What about dual native ISO of A7S3? It should be the main benefit. Not the low megapixel count.
Yes, I think that phenomenon is less a myth and more a tech limitation that has recently been ironed out. DSLRs suffered from this much more than the latest mirrorless offerings. For example, my 36mp Nikon D800 starts showing visible noise at around ISO 800, while my 16mp D4S never has noise at ISO 800, and only ever starts showing noise at around ISO 2500
There is also a possibility that at least in some instances in past people didn't scale images when comparing (they didn't upscale a smaller megapixel image to the size of a larger one), so they would view both low and high megapixel images at the pixel level, and when viewing them that way, of course higher megapixel camera will appear more noisy, since pixels are smaller, and thus the same number of pixels will cover smaller surface, and thus "gather" less light.
Great to see you both still making videos! It's been a while since I've been back in the game.
This realization hit me when I replaced my a7ii with an a7Rii. The “R” actually seemed to have better low light performance mainly because I could shoot at higher ISO’s and apply more aggressive noise reduction in post to compensate while losing very little detail in the bargain.
(Being a BSI sensor probably didn’t hurt). 😉
I got the A7 II and I'm bitter about all the good stuff that the R II has going for it that I don't get, like the lower noise, being able to charge while running, and subject tracking that actually works (on the A7 II it completely drifts away immediately).
@@404Anymouse I have the 7ii and ater 7 years of owning it i can say that it is a nice camera to learn, but the autofocus sucks, working it as a videocamera is a nightmare, but well is the second A7line, almost 10 years have passed since its launch, we cannot ask more from it
The a7Rii has a backlit sensor which helps significantly with low light performance, while the a7ii does not.
BSI-CMOS is about full f-stop better than CMOS and that's substantial difference
In regards to video, wasn't the A7Sii considered the king of low-light? Trying to remember why that was the case if dual-native iso was a luxury released with the A7Siii.
For photographers that need to get photos out fast, such as when covering sporting or other live events, they won’t be processing RAW images. They will be using the image quality of the immediate JPEG file from the camera. In that scenario, don’t the lower megapixel sensors produce better low-light images compared to the high megapixel sensors? Maybe that is what the camera companies are using for the claimed low light advantage of lower megapixel cameras.
I've never found this to be the case. Especially now that high ISO cameras can pump out downsampled JPEG images, there are really no advantages that I've found to having fewer pixels. My 26 megapixel images from the A7R V are still cleaner than 24 native megapixels from A7C.
Compare the jpegs from R6 and R5. R5 is stunning when it comes to Jpeg quality and details.
The 60mpx Sensor has strange color. You can clean it up nicely, but contrast and color fidelity seem to decline more, isn’t it?
This is the exact conundrum I am currently mulling over, thanks for shedding some light on this. Interesting analogy with the buckets, but surly the issue is not the total amount of water (light) collected by the total combined area, but the amount of water collect by each bucket? And therefor the need for less gain where more is collected? Not to mention the higher percentage of “dead” space at the perimeter of each bucket. It’s a lot to get your head around and I’m looking at this from the perspective of a layman, so I guess the results speak for themselves. Is the improved detail of the higher megapixel sensor not simply the result of higher resolution?
Was your test image map offset printed? If so it was a pattern of CYMK dots and not an ideal subject for evaluating noise. A better test subject would be smooth white and gray 3D objects in lighting with a cross lighting in a large black draped space (to eliminate spill fill from lights) with starting with no fill to create a lighting ratio ratio / contrast range which exceeds the range of the sensor, exposing so the only the specular highlights are clipping (255) and the smooth white surfaces have eye dropper readings of 250. The white object will be rendered white-to-middle grey and the grey object will be reproduced from grey to the noise floor of the camera.
The problem with your bucket analogy @ 4:40 is that digital cameras expose to keep the highlights from the highest area of reflectance in the scene below clipping. In high contrast scenes were luminance range is greater than sensor range the shutter will close and end the exposure before the “buckets” reproducing detailed shaded dark objects receive enough photos to record the detail so all that is amplified is the noise floor.
Physically larger pixel sites (higher volume buckets) take longer to “fill to the brim” with photons in the highlights which give those in the shaded areas more time to gather enough photons to register an image, why sensors with larger photo sites in theory should have longer dynamic ranges. I say “in theory” because it is impossible to know how the signals are being processed.
An empirical method I’ve used to test dynamic range and noise is to shoot a gray card in flat light filling the entire frame with a fast lens and start with lens wide open and a shutter speed which over-exposes the card to a point just under clipping (e.g. 250) you’d want to have a white non-specular highlight to have. Then bracket the exposures in one stop increments by first closing aperture and then doubling shutter speed if necessary until the spike the card creates on the histogram marches all the way from right to left. That test will measure the sensor range in f/stops and reveal via the spacing of the spikes how linear the sensor response is. The darker frames will reveal signal / noise level. Repeat test the for each ISO to evaluate how that variable affects exposure range, linearity of exposure and noise.
Thanks for the video and reminder that 'more is still more'. I had been considering an a7siii alongside my a7iv, but I'm minded to think that despite the higher cost that an a1 makes a better overall second body for a variety of reasons. Perhaps off topic but a pity that Sony's a1 flagship (which does an amazing amount) doesn't have a flip screen (trickier without a physical re-build - maybe in a1 mk ii) or firmware update features (it's prob just a firmware tweak needed) adding benefits like 'focus breathing'. Anyway for still and video it seems that an a1 is still streets ahead of an a7siii.
The A7siii does run forever, even shooting in 4K 120p 10-bit 4:2:2, so that's a plus.
A7 s3 is way better for video than a1
Why do you never talk about the Nikon z cameras ? I'd like to see comparison between Z6-2 and the higher mp cameras
I have both Z6 and Z7 (first versions). The same story. Z6 is apparently better but if you resample images to the same size and apply noise reduction, Z7 smokes out Z6 big time. Although Z7 has more noise, the image retains more details and is generally better, especially colors in shadows that gets very weird toward magent-ish tone on Z6 at high ISOs, while stay very nice controlled and neutral on Z7.
I've noticed that myself in several cases (less MP is cleaner, but more MP is more detailed so the final result is similar), but the idea that "less MP is better for low light" still lived in my mind. Thanks Tony & Chelsea for this revelation!
Even when you throw science at some people, they will still stick to their guns and claim the lower resolution camera produces cleaner images. I've argued with people about this for years, but it's like arguing with a wall. They simply don't understand the scaling bit. They really can't understand that a 50mp camera sensor can always produce a 24mp image, while a 24mp sensor camera can never produce a 50mp image. It's totally lost on some people.
The only true advantage of a smaller resolution sensor will be less work post processing to clean up high ISO images.
Please talk about another myth: Do lower pixel counts (on same sensor size) contribute to better dynamic range?
As you say, it is just a myth as well. Lower pixel counts (on same sensor size) wouldn't contribute to better dynamic range.
@@tubularificationed what is noiser, one huge basket collecting charge connected to one ADC or 4 smaller baskets connected to 4 ADC and averaged in software? Ofc it depend of ADC quality. If you get very good ADC it almost same, but in practice i very doubt. And ADC work in very hard conditions (crosstalk with adjucent lines/sensels, geometric tolerance of circuit elements must be 4 times finer, etc)
I'd say that 4 smaller baskets are privileged.
- tolerance errors are evening out better, if you have 4 small buckets rather than 1 big one.
- the same goes for dead pixels (or stuck pixels). Faults have only 25% effect rather than 100%. Repairing faults (e.g. interpolation between neighboring pixels) is also better, because at high resolutions, micro contrast edges are recorded more smoothly, which helps concealing mathematical interpolations.
- all demosaicking / debayering artifacts also benefit from high resolutions, for related reasons. Because also demosaicking is some sort of interpolation between neighboring pixels.
- 4 small pixels only accumulate 25% of electrical charges each, i.e. only would generate 25% of cross talk, and that would only affect 25% of the spatial neighborhood.
But in general, ADCs and other electronics are so properly designed (already since long), that such issues play zero role since many years anyway.
Which can be verified by inspecting any high-res photo taken during the last 10 years or so 😉 (or actually the entire era since when high-res sensors existed).
@@tubularificationed nice trolling, respect.
Really helpful video. A key point you mention is that some higher megapixel cameras skip pixels when filming because the processor isn't fast enough or cool enough to handle the light data. If we compare the FX3 (10.2 megapixels (effective) for movies) to the FX30 (20.1 megapixels (effective) for movies) do they use different processors ? Does the FX30 skip lines ? How does this relate to down sampling from 6 to 4 without pixel binning ?
It’s so fun to watch Tony when he’s needing out on camera tech! 😂 keep it up Tony! You’re doing GREAT!
I was just thinking should I sell my A7R4 for A74 due to the “poor performance” on low light on the A7R4, and this video just pops up on the recommend list on RUclips. Which helps a lot on my decision.
Now I have decided to keep my A7R4(which is good because the 2nd hand price of A74 is higher than A7R4 now), just need find another way to light up the image..
Glad I could help and sorry you almost got sucked into that myth!
I saw the method he used to get everyone confused, he scaled the high megapixel camera to 100% while scaling the lower megapixel camera to 400% but he never scaled the high megapixel camera to 400%
No he did scale it properly. The 12mp, image needs to be blown up more, because the computer thinks the photo is smaller then the 60mp photo. When in reality theyre both the same physical size.
I can't believe that in 2023 people still get confused by DPI/PPI and need absolute basics explained.
Though actually it makes sense... No one ever prints anymore, and those fancy phones of whatever resolution screens and cameras do everything to obscure that knowledge for marketing reasons.
@@Lishtenbird Maybe this person is just new? I dano... But yeah actually printing the photos, would be helpfull in learning this stuff.
@@LishtenbirdNo one ever prints anymore? News to me! I print all the time! How am I supposed to sell my images? NFT?
Detail and noise are 2 different things, we don’t want to see all the pores and bumps in the skin on a portrait, depends on how big you want to print your image.
Higher megapixel camera don’t have less noise because they have more megapixel, it’s because the signal to noise ratio is better due to the technological improvement in the firmware. However, they still are noisy at low light conditions when a single photo site did not receive enough light or low signal value, under same low light conditions, the lower megapixel cameras collect more light per photo site or better signal value relatively, hence cleaner pixel.
Even though 1 pixel in the low mp camera and 9 pixel in the high mp camera collect same light, for same total area, most of the pixel end up as noise in the high megapixel camera, whereas a quality pixel recorded in low mp camera.
Great, as always. Thanks. A few off topic comments perhaps for another episode:
- Sharpness may not be everything. So here is another myth: 'Fewer, larger pixels are far better for dynamic range than more, smaller pixels'.
- Moreover, for adapted lenses or more 'budget friendly' lenses (non professional) what is really the tradeoff between high megapixels and lower pixel pich?
Why are you comparing out-of-focus images for high ISO?
Bless you Tony! I went with a Nikon D800 for shooting live concerts. To the standard thinking of the time that is the opposite of what is recommended. The D800 is awesome in low light and allows for cropping a lot better than some other cameras.
I suspect there are additional factors to consider regarding some comparisons. (I am particularly thinking of Math Photographer’s videos on this subject). For instance, the Leica SL2 (47MP), SL2S (24MP), M10-R (40MP), M10 Monochrom (40MP) and M11 (60MP) have been compared in various combinations regarding low light performance. The SL2S is both marketed and regarded as better at low light performance than the SL2, but the SL2S also has a newer and different sensor (BSI). The M10-R and M10 Monochrom, released at the same time, have the same resolution but the M10-R has the Bayer filter and concomitant demosaicing engine. Of course, the Monochrom has no color noise, and also presumably gathers more light due to the lack of a color filter. The M11 has much higher resolution, but also several years’ newer technology, including physical changes to sensor architecture. None of this takes away from Tony’s premise regarding MP count and sensor area. I am just saying that not all comparisons are truly apples to apples (or Cosmic Crispy to Red Delicious).
Sorry Tony, I have to disagree here in regards to “real world” low light usage. I set up an ARs3 & an A7r4 next to each other with identical settings & lenses & shot a Milky Way Time lapse, & the A7s3’s images were much clearer & less noisy. I understand what you are saying in regards to scaling( 12mp v 61mp) but I’m just giving you a “real world” example. Lol keep up the great work. 👍
“Cleaner”, “less noisy”… but you forgot about the last mitigating factor: “more detailed”. Perhaps in your astro-photography scenario, with just point sources of light, the “more detailed” part is not as differentiated. Just a guess…. and something to consider. It’s an ‘unusual’ subject matter to be sure. Not like rendering cursive scripts from a map.
@@billr3053 when viewing as a 4k render, the extra detail of the r4 was unusable to where the noise couldn’t be removed to a satisfactory result, so for astro timelapses viewing in 4k, the s3 gave a cleaner result without detail loss.
@@nashhall694 Interesting. Astro photography is my weak point. I’ll take your word for it but as I said that “unusual” subject matter may emphasize other strengths of a larger sensel-laden surface. As opposed to a wall map whose EV has been cranked to 18% gray level unlike a black sky for example that is supposed to remain black. I just thought of that. :)
He doesn't define what he means by "BEST" and this is a good illustration. From his tests he obviously considers capturing finer details to be the issue at play, which.. like the sensor has 5x the resolution, of course it's going to capture finer details. But there are so many factors at play, and the RUclips incentive structure is at odds with the fact that there is no one definition of "best" when it comes to image quality. Is it detail? Dynamic range? Noise? Accurate color?
If one's objective is low light and high ISO, besides astrophotography being the perfect medium to test this in, one might also conclude that you'd be more interested in comparing noise or dynamic range (specifically the shadows). This video just.. it doesn't make any sense to me. It's trying to make a simple generalization about a multifactorial system based on a single factor. It's just doesn't demonstrate anything useful. Unless you consider showing that a higher resolution camera captures finer details than a lower resolution camera to be useful in some way.
Thermal noise is worse with high megapixel cameras. Your eyes are not lying to you!
I’m curious if this holds in *very* low light - like Milky Way photography under dark rural skies. I suppose I could compare my R7 to my R6 in crop mode. (But not anytime soon. Been a pretty wet winter.)
Thanks for the great video. I don't like to resize all my High MP pics, so I go for lower pixels. Just buy what fits your needs.
That's true. you just saved me alot of money
I wonder how does something like the Hasselblad X1DII or X2D compare vs full frame? they have high megapixel counts but have larger less dense photosites on the sensors. does that make for a cleaner image?
See I would like to know too. Does the logic of a larger “medium” sensor area reduce noise when compared to full frame if the MP are the same? Or is the result the same because the number of pixel are the same? Faster lens reduce noise...do large sensors?
If you are going to fiddle with noise reduction on every photo, then sure- I agree with you here. I shoot a lot of low-light PJ work and the noise processing in-camera for JPGs, for whatever reason, seems to perform better for me in the R6 than it did in my R7. I know it's also a crop sensor, but ultimately, I switched to the R6 from the R7 because it was far better in low light in my experience and the R7 was too fast for my needs. I can shoot 12,800 ISO on the R6 without significant issues, the R7 made it hard to shoot anything over 3200, and anything shot any higher than that certainly needed a heavy post-processing hand, as even out of camera JPGs were almost unusable.
A larger sensor will have better light gathering ability which is why the R6 works better for you
@@klausfuchs516why would that be the case?
Awesome video, I was really tensed about my a7rv purchase. Can you please make one comparison for ar7v 26mp vs 60mp image qauliity/noise comparison? Also, comparison to full frame vs apc video comparison?