Photogrammetry in Meshroom & Blender
HTML-код
- Опубликовано: 29 июл 2024
- Today we explore generating meshes inside of Meshroom and cleaning them up inside of Blender.
Downloading Meshroom 0:00
Alternatives to Meshroom 5:23
Launching Meshroom 7:58
Example Project 12:09
Brief Intro to Blender 20:46
Importing Mesh to Blender 27:42
Orienting Mesh 29:08
Cleanup 38:15
Follow me on Social Media for more art, tutorials, and announcements:
- Instagram: / crompwell
- Twitter: / crompwell
Leave a comment below.
Please consider supporting the channel!
Thanks for Watching!
ko-fi.com/crompwell
paypal.me/kcrompwell
This was everything I needed to get started with photogrammetry. Goes to show that a good video truly is timeless! Thank you, and good job
This is the most friendly and easy to follow and engaging person in whole world of RUclips. Amazing I just like every idea expressed.. Is it just me
Thanks for this. By far the most clear explanation I have seen so far.
Wow this is the best tutorial i've seen so far for learning blender - you have a new sub! You surely deserve way more than 413 subs - Really good stuff. Thank you!
Best tutorial I found for beginners, all the "wasted time" on Blender really help a lot.
Glad it helped!
Thank you so much for this brilliant and concise tutorial on how to use these free and very powerful tools. I have struggled with Blender in the past, but still managed to achieve the results i wanted (albeit frustratingly). Your tutorial has switched a light bulb on in my head! As for Meshroom, wow! What a fantastic piece of software. Now i'll never need to make a mesh from scratch again....possibly ; ) Thanks from the UK.
Thank you very much for this great video tutorial. I will learn Blender (I have the book) and then I will start also with photogrammetry. Thanks! Martina (53 years)
I'm glad you find it useful. :)
Thank you posting this! I've been playing around with Blender since version 2.79, but just started working with photogrammetry. I found reference to Meshroom and Agisoft De-Light in a book that I recently purchased. So of course, I had to hit the net and find out everything I could. Wow! A brand-new world! Thanks Again! 😃
Glad to be of help :)
Thank you so much for your tutorial - you covered so many basic questions I just wanted/needed to know!
this is really helpful. You explain everything so well! thanks
BEST TUTORIAL SO FAR.........THANKS A LOT.
Great tutorial mate, thanks a million!
Really great Tutorial, I needed that :-)
clear explanation for beginner
Great tutorial! Thank you very much!
Excellent! Thank you for very informative and useful video. Just made my first 'copy/paste' object - wooden vase into 3D printing file. Right now printing scaled down model. Good details... okay, here and there are little errors, but nothing serious. You have new subscriber. Anxiously waiting for more.
This is very interesting!! Thanks for sharing!!
Thank you so much. Great tutorial!
Glad to help :)
Excellent tutorial.
Thank you for helping others.
TechsObserver I try
Nice video. In blender you can rotate when in side view by clicking on the white circle gizmo (the view's rotation and since you are in side view then the rotation will be rotating along X) Just an FYI. Or you can hit r on the keyboard like you said
Thank So much for your video.
No problem! :)
Awesome thank you so helpful
Wonderful, excellent tutorial. I recently bought a drone and want to get into drone mapping. But I didn't want to spend the big $$ and buy pix4D or other expensive software. This really helps me out. Also I've taken blender classes before. Thank you.
I appreciate the kind words, and I’m glad you found it helpful :)
@@Crompwell I forgot to ask, do you know of a free or reasonably priced alternative to sketchfab for showing our models. Something that would allow you to embed the link to a website.
Fina Bentley-Kimura A word of caution about uploading 3D models to the web: with platforms like Sketchfab and WebGL, there is always a way to reverse engineer anything that gets downloaded to computers for user-side interaction. I’ve seen sketchy people offering to rip 3D models from sites like Sketchfab for a fee, effectively breaking Sketchfab’s market element. If you’re planning to use Web-based 3D for commercial purposes, I’d highly recommend avoiding publishing the actual project and use videos and gifs instead, to avoid the thieves using 3D model ripping scripts.
If the commercial aspect isn’t important to you, then read on:
Apart from Sketchfab’s limited free account, there is WebGL which you could use to show models off, but it’d require you to program it to work.
There is a Blender to WebGL addon called Blend4Web. The base version is free, but kind of limited AFAIK. It also only supports version 2.79 instead of the newer 2.8x Blender builds. I’m not sure if it’s been well updated.
I believe you could also use Raylib, a free and open source programming library that can handle 3D graphics, but I’m not sure how well that’d integrate with the web. It’s probably doable with this library.
I’d say that WebGL is your best bet and there are some tutorials out there on Blender to WebGL.
Fina Bentley-Kimura Verge 3D also just released integration with Blender to get Blender scenes working in a web browser. Not sure if it’s free or how it works.
I don't even need this entire tutorial, but I kept it running in the background because your voice is so incredibly calming
Haha, glad you enjoy it.
Thank you very much !!!!
how would you get the texture onto a uv map? would you have to bake it onto a retopod model?
Yes, you will want to bake your texture onto a UV retopo'd model. "Precompiled" is a folder I made. Everything inside of it is what you will find wherever you saved your project.
I noticed that Meshroom was updated to a 2019 version. I will check it out and post an update if there are significant changes.
Wish my attempts at photogrametry were as good as that to start with....will keep plugging away with Meshroom hoping to crack it. :-D
Maybe know this by now. Use middle mouse button to drag point set into the origin of the scene. You can move your focus to the center, by dragging it. I know its an older video. Still great. Thank you!
Man, I wish you'd do an updated video on this again...my screens look nothing alike...I stumbled my way through this and got the mesh trimmed and exported but, it wasn't without flaw. Aligning the axis' and gettin things sorted was tough.
I’ll see what I can do.
感謝教導
Thank you teach me.
I'm fron Taiwan.
Glad to help
This tutorial is fantastic! Thanks for taking the time to help us all out. By any chance though, have you ever been unable to drag and drop into meshroom? I keep getting this issue
Some of my students have run into this. If you are trying to drag photos out of a google drive or zip archive you can run into problems. Make sure the photos are store locally on your PC, then drag them from the file explorer and into meshroom. Then it should work.
How do I get Meshroom to make a obj? I finished the scan and it's done but no obj file is made in the cache folder for the project.
If Meshroom has finished the whole process, your obj will be located within the meshroomcache folder > texturing > folder with long name of letters and numbers > obj file with mtl + texture pngs + a few meshroom specific files.
Hmmm, that model reminds me of Mickey Dolenz's creation 'Metal Mickey' robot in the late 70s TV show in the UK. 😊
stevekitt52 I’m not familiar with that, but I see what you mean :)
Hello Crompwell, thank you for the great tutorial. I have a question, is there a way to create a video of the finished model? nothing fancy, just a simple video rotating around the model for sharing online in an MP4/AVI etc format. Thank You!
You could set up a simple turntable video in blender and render it out. Check out my latest video on cameras, it’ll show you how to set that up. ruclips.net/video/p7AU3NkKhJY/видео.html
Thanks 😊
Thanks for this tuto on meshroom & blender :)
When you take the pictures, don't you think it would be better to turn the object instead of moving around? (this way the light is the same on all the pictures)
No, Meshroom uses structure from motion, which means it uses the background and camera information to determine the object's relative position in space, then reconstructs the object based on the data that is present.
If you want to keep the camera stationary, you will need different circumstances to satisfy Meshroom's structure from motion requirements in order to successfully rebuild the object.
In that case, in order to reconstruct the structure from motion, you would likely need to take many more photographs and to ensure that the background pixels always remained the same in each shot. This way the only pixel motion that would be considered would be the movement of the object itself. A green screen or a colored background may work, however these circumstances may also require automated background removal and tweaking the settings in the Meshroom nodes. I'm confident it can be done, but it's a different workflow that requires more work.
You can also install ubuntu linux on a mac with nvidia gpu as I did. Linux can read apple partitions and runs with much less resources than osx on the same hardware.
Yes, but I’m not sure how many users want or know how to do all that.
@@Crompwell Honestly installing it on linux is not as simple as it would be somewhere else. Modifying the path to add variables and so on. Many Mac users are developpers and I think they should consider linux instead of osx because Apple is slowly getting away from intel chips and will for sure give up their macs in a few years.(ARM os updates too expensive to adapt to non-ARM chips)
@@zoranvelijevic7871 I agree with you in terms of getting away from Apple due to their practices and components. That said, there are a good number of non-developers that like Apple because of the simplicity of the OS, despite my attempts to convince them otherwise, unfortunately.
@@Crompwell I agree, but it's also important to explain to people that linux can be installed instead of osx because apple's hardware is in fact very standard and doesn't make any problem regarding drivers.
The Arm OS will of course be different but the future pcbs will for sure embed common sound and video chips, overpriced of course because apple buyers will keep buying low quality hardware for a very high price.
You'll need to update this sometime to Blender 2.8 and the current version of Meshroom.
David Thrasher I’ll see about making a new video. Meshroom 2020 should be on the horizon if not already.
I'd be interested to know if there is a way to project the original texture onto the decimated model in Blender. I know there is a workflow in ZBrush where you can generate normals & bump map from the hi-poly version onto the low-poly version, and even completely retopologize the model & UV's automatically to create a more readable texture file.
Yes, it can be done, however, you will need to unwrap the decimated mesh. This can be done with auto-uv, but it may produce wonky results. The baking can be done with cycles renderer inside of blender or with an external baking engine.
thanks
still didnt get a good output but realised i was my house lights and also camera phone was only 2mega pixels needs to be 12mega pixels for photogram
hi, i'm usieng mehroom and the process going well to the last node, but the mesh button on the 3d view won't work and the color is grey not white? what should i do?? is there any mistake? thankyou.
If Meshroom was able to complete the process (the green bar made it all the way to the end), you will find the textures and model in your project folder > MeshroomCache > Texturing > [ long string of numbers file ]. If this is the case, you should be able to hit the mesh button, however, I have experienced moving the file and losing the option to mesh it again in, however the model and textures we're still in the texturing folder.
If the process did not finish, you will need to see where it stopped and you may need to delete the folders in your MeshroomCache after the point where it stopped. (E.g. If it stopped on Meshing, look in the MeshroomCache folder and delete the contents of the Meshing folder as well as all the contents in the folders corresponding to the nodes that come after Meshing).
Then, start the process again. It should show the green bar wherever your process successfully completed and should finish without a problem.
A word of caution. You should save your project before starting the process and you should work locally on your machine, not on external hard drives or SD cards. This includes any photos you use in Meshroom. Make sure all the files on are your computer and are not being accessed on an external drive
Is Nexus free for soft soft
Thanks for this dopest video! I have some Question.. Can i get a low polly of photoscan of buidings and room ! The problem is can't optimize buildings it have 3500000 poligons,and its unreal to optimize in Unreal Engine 4 in game development. Howe can i get optimize? Ore i can make low polly in Mash Room? Thanks
Алексей Брижа there are some nodes and settings within meshroom that can help reduce the polys, but I would suggest building your room in parts rather than all at once.
What kind of tolerances can you expect with meshroom, and how susceptible is it to the stacking effect for large objects such as a mid size car?
I can't speak to the technicality of your question, but I do know that Meshroom can handle larger scene sizes.
Alicevision, the company that has made Meshroom, has a few examples on their Sketchfab page of landmarks that appear to have captured some decent detail given that they were captured on an iPhone. These examples are a few years old, and likely make use of a dated version of Meshroom as well (there is a fresh 2019 version out now).
The only scene I have tried that may be comparable to what you're asking is this: skfb.ly/6IVsT
That was done hastily, without a tripod, and with about 35 photographs, shot with a Canon 80D. I'm confident that if more time was invested, with additional photos, the reconstruction would work quite well.
A General rule of thumb that I give my students is to have 70-80% overlap between pictures, regardless of software. The bigger trouble with doing photogrammetry on a car will be with the metallic/reflective paint and translucent surfaces.
I have tried to use meshroom but not very succesfully, of the 35pictures submitted on 2 had ticks, the rest were minus, I have checked my invidia card and its a geforce MX 150 in a HP - H47OURBV laptop, cant find it in Nvidia site and in the old geforce section to see if the computing power is above 2.
This may indicate that your card isn't supported. The lack of checks on your images may mean that there is an issue with the quality of you images. Ensure that they aren't blurry and that they contain enough of the image for the software to recognize from picture to picture. Using a tripod and steady, color calibrated lighting will also help.
When I'm in edit mode, I select the part I want to delete (like at 39:26) and delete the vertecies but there is still most of the selected party that remains... Any sulutions (I'm using Blender 2.8)
Make sure that you are selecting in wireframe or Xray mode and make sure that you're deleting vertices. If you delete by vertices, it should remove everything selected. If not, I'd have to see it to tell you more.
Please help me, i cannot see the nodes in my graph editor, my nvidia is geforce gt 520mx (2.1 compute capability)
Send me a screenshot.
hello i have a question: how good my camera needs to be? , i am talking about the lens distortion, does that affect the result? if so, should i use a DSLR with a lens that's aiming for geometry accrucy ? what about focal length? the longer the better or shorter?
thanks.
Most modern cameras should be sufficient to produce a usable mesh, many of my students even use mobile phone cameras, however, a general rule is that a better camera will produce a better result. I have seen that full-frame sensor cameras provide very good results.
Photogrammetry actually takes lens distortion into consideration in its calculations so you don’t need to worry about this and you should NOT correct it in post, the software actually requires it.
As for focal length/depth of field, you will want to have photos with as much clear and in focus detail as possible. The more clear and non-blurry data you can provide the photogrammetry software, the better your results will be. Photogrammetry software uses both foreground subject and background detail to calculate structure from motion.
I hope that helps.
Thank you for your video.
I have a problem with mine, I put the photos in the program and it only renders a single image.
I can't select all of them, why does this happen?
I'm not entirely sure. It could be that Meshroom is rejecting your images. Perhaps there is blur in them.
It could be the naming convention used in the images. Be sure that they're all on the local disk.
A lot of people have problems with images not being on the local disk and were instead trying to work off of external storage. That is the fast track lane to headache land.
Great job! So what video refers to convert 3D to greyscale map depth in blender?
No, although you could do that. This is about a software that can approximate the surface and volume of a 3D object by extracting data from the 2D pixels and their movement across a series of still photographs and converts it into 3D polygons/geometry for use in applications like blender.
Great video, very informative ! Could you comment on the possibility for measuring the 3-D objects (linear, angular, volumetric) ?
The only way this is possible is to include some kind of metric in the scanned data. You’ll need some kind of reference to determine the measurements of what you capture since the 3D data is generated by structure which is reconstructed based on movement. So, you could probably use the camera’s sensor width as a useful metric, but it’d be better to capture something directly like a rule in your photography, for example.
@@Crompwell Thank you ! Suppose I take the photos with a measuring tape placed directly on the object and then use that to calibrate the scale, could the measurements be done in Blender ?
@@maxfacdoc As long as your photos are clear and the reconstruction goes well, I don’t see any reason you wouldn’t be able to do that. :)
This is an inspiring start for everyone! What camera and settings are used to capture the Details? Trying to decide what camera to use for capturing finer details. Thanks.
If I recall correctly, I used a Canon EOS 80D. You can get pretty good results with modern camera phones, but a professional DSLR with good quality and proper lighting will give you the best results.
@@Crompwell yes 20mp plus is the way to go, thanks for confirming. My cellphone is 12mp and scans were too lumpy. Just bought a used Nikon D800 from keh, 36mp. Appreciate your clarity before such a purchase invesment.
Do you know if it is possible to use Mushroom on a bunch of digital pictures... pictures without shadows or backgrounds? I have 60 images of a 3d rendering from every angle and I would like to convert it back into a 3d rendering. Thanks
It's possible, but you might run into an issue with the camera database being able to identify the sensor data since they're 3D renderings. But I have heard of people using it this way.
What were your system specs on this video? I have a 1050ti and on the "Depth Map" part it takes up to 40 mins, as you say on the video. So I wonder...is that something about system specs or just the process itself?
At the time of the recording, I believe I was using a 1080 Ti. That said, the amount of data will change how much time is required for processing. 40 images @ 72 ppi will take significantly less time to process than 200 images @ 300 ppi.
AT 20:18 you mentioned making a blender file from Meshroom to import/open the file in Blender. How does one make that happen? There is also not a "Load Model" button. I am using Meshroom V 2021.1.0
Once you’ve finished the meshing process in meshroom, the “load model” button will show up. If you haven’t completed the process, there won’t be a model to load and the button won’t appear. Once it does, you can export the model as an obj and import it into Blender and save out a blend file.
Hello good people, i need to scan an engine for my project and i wonder how dimensionally accurate the 3D model from this method will be. Is there a way to specific dimensions to make sure that all sizes are correct? Thank you in advance!
ruclips.net/video/b1t33KcBlnk/видео.html
i had issues with red flags...turns out it was my anti virus software blocking the app...once disengaged it functioned again worth noting that is all
Sweet video mate,
I'm stuck at 44:00 when I try to select my whole model it only selects the faces which can be seen from the current view. This means that when I invert the selection is also selects faces I do not want to delete (on the back of the view from which I've selected). Can't seem to find the right button must have some setting activated which should not be. Any tips?
Press the Z key before making the selection in order to turn on wireframe mode. This will allow you to select through the model.
Thanks a lot, worked like a charm. Thanks a lot for your super clear instructions. You don't skip any steps and repeat all the thing I need to remember in the beginning stages. Are you teaching a subject at a university or what?
dries v Glad to help. Yes, I teach this at a local university.
can we deal with pointclouds & depth images in meshroom using a stereo camera.
I'm not sure. That's something I'd have to look into. You could try taking a look at their web page or github and see if they have anything about that.
If you close the command console during the pipeline process meshroom would close and you would have to start a new project since the last one would be bugged out. Fun accident. First time I used it I thought I could just close it like the blender splash screen.
I would like to learn this photogammetry for a reason of restoring old equipment of rather larger dimensions. I need to create point cloud or mesh, convert it to surfaces or solid, usable by the CAD program. Then I will re-create the part inside of CAD by approximating the model, so I get workable geometry. These technical parts have a lot less details.
After listening and watching your and some other videos, it is not quite clear to me, why do I need Blender. Can you point me to some other source, where these two programs abilities and usefulness is explained? My objective is to get surface or solid model. Thank you.
Blender is just the program that we are using to manipulate the obj mesh that is being exported by Meshroom. Any standard 3D DCC will do for this. If you want to export a point cloud and manipulate it in some other program, you're more than welcome to do so.
Hey guys, i have just tried meshroom today and i had some sortof succesfull results. But now i have a finished a project, but for some reason the mesh option is grey and does not show. Does this mean that it could make a mesh? and if so, why do you think that happened?
If Meshroom was able to complete the process (the green bar made it all the way to the end), you will find the textures and model in your project folder > MeshroomCache > Texturing > [ long string of numbers file ]. If this is the case, you should be able to hit the mesh button, however, I have experienced moving the file and losing the option to mesh it again in, however the model and textures we're still in the texturing folder.
If the process did not finish, you will need to see where it stopped and you may need to delete the folders in your MeshroomCache after the point where it stopped. (E.g. If it stopped on Meshing, look in the MeshroomCache folder and delete the contents of the Meshing folder as well as all the contents in the folders corresponding to the nodes that come after Meshing). Then, start the process again. It should show the green bar wherever your process successfully completed and should finish without a problem.
A word of caution. You should save your project before starting the process and you should work locally on your machine, not on external hard drives or SD cards. This includes any photos you use in Meshroom. Make sure all the files on are your computer and are not being accessed on an external drive
Hi, I'm having that issue you mention in the Depth map section. It becomes red and when I hit start again it doesn't continue, but returns to red and stops processing. What should I do in this instance?
If you have a CUDA enabled GPU with updated drivers, you can try deleting all the folders including and after the depth map node (from your project folder) and open and run Meshroom again.
Hello everyone me and some colleaugues are starting the 3d scanning of a looot of the Italian heritage (like statues and so on) to make disponible 3d models identical to the original for games movies an so on, do you think that somebody will be interested?
Is it complicated to reproject the image on the decimated model?
That depends on the complexity of the model. I have a more recent tutorial that shows how you might go about doing this.
#thumbs.up.
Mine totally went in the red in depthmap i keep hitting start again and it does not work. The green start just turns green again
Sorry to hear you're having trouble. The command prompt behind Meshroom should give you an indication of what is going wrong. Depthmap is the node that takes the longest to process and tends to fail quite a bit.
Thank you for this tutorial! But my imported .obj file does not show the material in Cycles Render, only in Blender Render. Does any body have had the same issue?
First issue is that you are using Blender version 2.79, which I know is what I used at the time of this recording, but if you use 2.8 and above, it will work much better, and I do have more up to date tutorials which cover this, just check my channel for the photogrammetry retopology videos.
Secondly, IF you are choosing to keep using 2.79, make sure that you switch the render engine to cycles first, before importing your obj, otherwise it will only work in Blender's internal renderer. I make note of that in this tutorial at 27:23
@@Crompwell Thanks again! I already tried switching to Cycles Render before importing, but no luck. I wanted to migrate to 2.8 anyway, so that comes in handy.
@@MrNucleosome Hmm, in that case, the only other thing I can ask you to check is that the PNG textures are present in the texturing folder and that your obj and mtl are present there, as well. If these files aren't present, Blender won't know where to read them from when you import the obj. I think this will be the same in both 2.79 and 2.8+.
I wish you all the best with your migration to 2.8+. I highly recommend it. It's leaps and bounds better than 2.79 imho, and makes a lot of things easier.
Has anyone a idee why meshroom at the point "depth map" stops and can't go further? I've tried many different photo sets, but it's always the same problem
Monkey Dude What kind of GPU are you using?
I am trying to photoscan leatherite sofa in meshroom but the gloss makes the sofa look like melting ice cream and the chrome legs are squished pipes. How can i remove gloss and highlights in photoshop to get a better final mesh?
That can be solved by making sure you light is more even distributed across your subject. Using a diffuser may help. Otherwise, you may need to coat the sofa with something to make it more matte. The reflectivity of the sofa is causing you problems.
@@Crompwell
Yeah I know but I corrected exposure in camera raw and results improved in meshroom.
also I had cleared noise as well.
Now its metallic legs have so much reflections of surrounding and white light that the appear clumps of point cloud in meshroom and chewed straws in blender.
That is why i was looking for retouching workflow that could better my results in meshroom itself.
@@abhirutmeherishi5738 You can retouch things in post, you just have to make sure you're doing so accurately.
For things that are metallic or just otherwise not good for photogrammetry, you can create those in 3D, from scratch.
how did you open the meshroom file in blender?
In blender go to file > import > wavefront (.obj) and click. Then in the file navigator, locate the obj file in your Meshroom cache folder. It should be inside of the folder structure in the texturing folder.
how many photos are the max that you can import in this free software.. my computor lost signal and crashed.. this never happened before. i am wondering if i attempted to many photos.. 148 pictures .this is what i was attempting
148 photos should be fine, but I have seen some students' machines crash due to naming convention (make sure the file names don't begin with underscores) or trying to work off of an external drive instead of locally on their machine.
If I have AMD graphic card can I download meshroom ?
You can download it, but you won't be able to use it without an Nvidia graphics card.
Ok so I am not getting responses from other videos I have watched online. There is not a ton of info on this. I got into blender about 3 months ago. I am a videographer so I have a nice camera. I am shooting on Sony a7riii a 42mp camera. Will I gain much from shooting these raw? The amount of conversion time just to raw to Tiff is time comsuming and the processing on these files is huge. Also, what is a sweet spot for amount of photos on objects? I know this will massively differ depending on complexity. Is just shooting these jpg sufficient?
In my experience, using JPEG or PNG will allow you to process your 3D model much faster due to smaller file size. I often recommend JPEG to my students and it works well. I'd keep a backup of your RAW files if you need to make edits or project texture detail, later on.
When doing photogrammetry, RAW simply takes too long to process due to the massive amount of data you're throwing at the application.
I have gotten reasonably good reconstructions from a minimum of around 30-35 photos, but ideally, if you want to have a detailed model, you can shoot upwards of 200+ photos. I'd say anything between 75-120 photos is a decent range. This will vary depending on how much detail you need and the size of your subject. Keep in mind that you'll want to have about 70-80% overlap of detail on the subject between shots.
Meshroom does have functionality for adding in additional photos after your first batch of processing, so long as you are shooting under the same conditions as the first session. So, if you need more information than 120 photos will provide, you can add more photos to the reconstruction AS LONG AS you can take them under the same lighting conditions.
Honestly, anything in the 12-14+ MP range works well for photogrammetry. Someone did a comparison between camera phones and high res DSLRs and found that sometimes lower res photos capture the overall form more faithfully, while higher resolution photos can produce some error in the detail.
It's not always this way, and it is true that the more good data you have, the more accurate your reconstruction will be, but the key is providing the application good data. Evenly lit/diffused lighting shots with non-reflective surfaces that have consistent pixels which help the application determine the structure from motion is key. Again, It may be worth using lower resolution shots to reconstruct the overall form and using high resolution shots to project finer detail.
I hope this helps.
@@Crompwell thanks so much. I am running my second test this time using convesions to jpg. I am running a rtx 2060 on my laptop and it takes hours. I am assuming the amount of pixel data this thing has heavily increases the process time? For this batch I did about 70 ish photos. And I have been shooting at sunset in shaded areas for a nice even light all manual. I do have an a6500 that is only 24 megapixels. Would this be the amount of resolution of even the jpgs or is the rtx 2060 just a little slow for this?
@@jordanruzic3665 The larger/more dense the photos, the longer they take to process. 24MP jpegs should work just fine, 2060 should also work well but might be a bit slow on laptop GPU.
@@Crompwell thanks so much for the tips. It has been hard to get some! I will switch over to the ol trusty a6500!
i got a problem , if you don`t save the file in the same disk as meshroom it does not work , you need to save the meshroom file in the same disk ass the app located
Yes, everything should be on your local disk. You should not be working off of removable disks or memory cards.
It's a damn shame that Meshroom only supports Nvidia GPUs. They could have made it way more useful if they supported OpenCL too.
Hi. I've finally managed to get through the Meshroom stage! When I press 'Load Model' and the check at SfM there is no smooth image as in your video.
When openiing .ojb in Blender, it looks like 'extra' bits load and not the actual image of the photos I took. Any ideas? Thanks
Sounds like it might not have reconstructed well. It’s really hard for me to know without seeing what you’re seeing. Make sure your photos are clear and free of camera shake or blur. When the OBJ is loaded in blender, it should load the 3D model with the texture from your images.
Thanks for quick reply! Think it might be a problem with my RAM :-/
I'm running 6GB and using Meshroom 2018 (2020 doesn't get through DepthMap stage).
Frustrating!!!....
@@davyp3118 Ah, that could definitely do it. I have run into problems with RAM limitations when using meshroom.
@@Crompwell
Seem to have managed to make enough space to get through most of the project (84 images) until the final stage 'texturing' where it just waits and does mothing. Tried altering settings in the texturing node but no luck. There is an .obj (12,500 kb?) in the texturing cache folder and tried opening in Meshlab but shows as a flat image...nothing like 3D. Any ides? Thanks
@@davyp3118 I can’t be sure, but you might need more vram to save the image texture(s).
I cant get recap to render my complete model, it only does one side. which brings me here.
Well, you can try to follow this tutorial to have Meshroom create the mesh for you. It may have to do with your photos and how you captured them.
Meshroom work only in nvidia cuda card! that why in depth problem.
He made that VERY clear - several times!
Thank you for this video and very clear explanation. I am a 55 y.o. entrepreneur in Toronto canada and I am inqiuring if you do an online course for both meshroom and blender but in more detail. Please let me know what options there may be.
rob sunderland I appreciate the kind worlds! I am in the process of recording and releasing a more in depth series on Blender itself.
After this, I will also be working on a more in depth photogrammetry course that will likely focus on Meshroom and a few other photoscanning software alternatives.
Is there anything you’d find particularly helpful for me to focus on?
@@Crompwell when do you think you will be able to release the blender project. What version of blender are you using for the upcoming blender project. It would most helpful in acquiring information on editing a project and sending it to a company like sletchfab to have printed...
rob sunderland I will hopefully have it released before mid April. The version I’m working with is the latest stable version: 2.82a but it should work for any normal install of Blender 2.80 and above. I can focus on showing you how to optimize models for 3D printing. As far as I understand, Sketchfab doesn’t offer 3D printing services, just online upload, viewing, and download. Shapeways, on the other hand, does offer 3D printing services.
hi,i try to use this app on my old notebook with gpu geforce 610m, cpu i3 and 4gb ram,i take photo using my realme narzo cam,and i already add it on cam database(but it still can't display the sensor width),i try to hit start button,some my project has 50 pic it go until got error on prepare dense scene,but if i try on project that has 100pic it freze on step featurextraction,whats wrong,do my notebook can't handle or my phone camera bad?
This may mean that you gpu is not compatible
@@Crompwell but i alredy check my gpu support cuda 2** but i don't know if my driver support on win 10pro
@@tjiansukarto8200 You will want to ensure that your graphics drivers are up to date and that your computer meets the following new requirements: NVIDIA CUDA-enabled GPU (built with cuda-10 compatible with compute capability 3.0 to 7.5)
can i use meshroom for bigger buildings ?
Soldier87xxx as long as you can get a clear enough image/usable data, yes!
@@Crompwell ty i will try it xD hope my computer dont die ^^
I have mushroom on my computer and I've been trying to get it to work.just to make sure it wasn't me I downloaded someone else's photos and watch them use them anytime I try to drag and drop all the photos into the left side of the screen that won't accept them
This may indicate that you are having an issue with Meshroom's compatibility with your hardware. Ensure that you are working directly off of your computer and not off of external devices such as portable hard drives.
You are sure that this photo-set works inside of Meshroom, elsewhere?
@@Crompwell hi, hardware wise i have nvidia 1060, AMD fix 8350, an ssd and hard drive. 16gb o ram.
@@lastchance848 I would imagine that the 1060 would work, but the AMD card will not.
@@Crompwell the AMD part is my CPU. I can look at pictures in the image viewer I just can't bring in like a folder to use to render that far left side I won't let me drag and drop anything into it
Is there any way you can send me these images for me to test? @@lastchance848
there a meshroom manual in progress... stills lacks a lot of info but it's a WIP
you can plugin a 'meshResampling' node between the Meshfiltering and Texturing nodes (add meshresampling, duplicate your already calculated Texturing node, remove link between meshfilterng and the texturing copy, insert meshresampling node between these two, calculate the second texturing node, and you will probably get a 50% polygon discount with same texture details, and because it's a separate branch of the computing tree it automatically creates a new cache folder
Hi! I am working on a model and after decimating to aroung 26.5k vertices my computer lags terribly when doing some sculpting. Is 26.5k still too many? I am working on a razer laptop with an rtx 2060. Thanks again and great video!
Is this after applying the decimate modifier?
@@Crompwell yes, original had about 1 million vertices.
@@amitzahirr I see. What kind of cpu is in your machine?
@@Crompwell 9th gen i7
@@amitzahirr That is bizarre. That should be more than enough to handle 26.5K poly's. Sounds like something isn't right.
I want to make sure that you have applied the decimate modifier, because if you only add it to the object without applying, the object will still have all the original geometry, even though object mode will show a preview of what the modifier will do after it has been applied.
Alternatively, you could try the Blender experimental sculpting branch which has been optimized to handle more geometry.
I have 95 pictures in my project, but about half of them has a red camera in the corner and if you hover the mouse over it, it will say: camera not reconstrcuted. What to do?
This means that meshroom in unable to reconstruct that camera angle. It could mean a few things: blurry photo, poor lighting conditions (too much contrast, too much shadow, highlight etc), inconsistency in the background, unexpected specular highlight on the surface of the object, etc.
The only way to avoid this is to make sure you have good control over your lighting and environment and shoot non-blurry photos.
@@Crompwell Thank you - will try!
Thanks for this video. why does meshroom reorder my photos out of sequence and only recreate the front of the object?
I am not entirely certain why it would do that, but it could be that it is rejecting some of them and can only use a few for the reconstruction.
@@Crompwell it seems that my "professional" shots using a light box and editing them in lightroom for clarity, contrast, etc...is not what it wants. Better results from slapping the object on a stool is a reasonably well lit room and moving around it with a handheld camera works better.
@@-thirteen That is very strange. Did you use two different cameras? A discrepancy between two different sensor widths could be causing issues.
@@Crompwell same camera. Just different rooms and lighting. Also no post processing
@@-thirteen That’s so strange. I’d make sure to check for any blurriness and if there’s none, try only using the “professional” photos and see what kind of result that gives.
what resolution is your camera? take note you used natural light it does not work with led house lights due to contrast and glare issues and also it cannot work by revolving the subject matter only by working around the subject, tried it first time saturday so need to play more,
Diffused/low-contrast lighting works best. Also keep in mind that reflective or shiny surfaces don’t work as well as matte or diffuse surfaces. I have used cameras with 12-24.2 Megapixels.
@@Crompwell exactly what i found a shiney box gave reflectance added to the glare from roof lights stopped it working but i think if i get a high res camera around 15 megapixels for doing a tall ship which is all matt should work better...down side ships is in water ...that is the reason i didnt laser scan the outside and only inside ship. Hoped photogram may work but i got a suspicion it wont especially as subject is huge...any thoughts?
@@dominicrobertdillonbrown9938 You may be able to reduce glare with a lens polarizer. I haven’t had to do this much, personally, but might be worth looking into before you go for expensive scanning alternative.
@@Crompwell already own a lidar scanner that is how i did the full interior of the ship...my only reservation about photogram is same as scanning ship moves up and down relative to dockside a few inches in tidal swell...i tried scanning outside of ship sadly got a wavey sinusoidal result...this maybe algorithms may struggle with feature definition in similar way
@@dominicrobertdillonbrown9938 Ah, yeah, the only other thing I can think of, if money isn’t an issue, would be to get a bunch of drones coordinated outside the ship and have them either simultaneously photograph or scan it. That’d be a massive and expensive undertaking, but I imagine something like that could work. Otherwise, you might have to capture what data you can and reconstruct the errors manually from measurements you have.
I dont know if this was mentioned or not, but just to add an important detail: Meshroom runs off CUDA cores, which means you need an Nvidia card to use it. People have tried using it with AMD cards and it practically just crashes.
Yes, I make mention of that at 2:21 in the video.
@@Crompwell ah makes sense, I was speeding through the video and didn't know it was mentioned. Great, comprehensive tutorial for beginners!
@@Vengehood :)
Meshroom wont work easily for me. I always stop and if I dont stop on a point the scan will turn out horrible.
Saturn What do you mean by “stop on a point?”
Where do I find my .obj?
27:42 please watch the video, I explain it there.
@@Crompwell alright
@@Crompwell oh ok thx!
Sorry for interrupting you agin, but I have a problem again now :| My problem is that my object has holes in it. Do you have an idea of how to fix that?
I cover a method for that in my more recent photogrammetry videos.
@@Crompwell Thanks!
Too bad that version 2020 is completely different and I can't find the Load Mesh button -.-
It’s not really that different. Once you get to the end of the texturing now the load mesh button should appear in the 3D window.
Mac has photogrammetry built in now.
Some IOS devices do, now. I believe.
You can copy the object, decimate it and then project the texture form the original model onto the lower res model.
This removes all the texture jaggies and should preserve the texture completely.
I think I saw this in a tutorial on using magicavoxel.
||ruclips.net/video/5MY3rsq5JGw/видео.html|| this one. I found it.
Hmmm my model comes out very rough.....extremely rough to be honest
Surprisingly, that can happen when you use a high quality camera. I'm finding that some camera phones will, at times, produce smoother mesh results (although less detailed image textures).
There are some settings you can mess around with inside the meshroom nodes, but personally, I prefer using a smoothing node inside of blender afterwards.
Now, if it's rough in an unusable way, like there are massive chunks missing from the model or holes where they shouldn't be, that could be an issue of lighting or the surface material (over exposed highlights/dark shadows or reflections) of your object.
What happens if I don't have a card?
If you don’t have a cuda enabled gpu, the application will have an error and you won’t get a result.
@@Crompwell how do I fix this just using cpu?
@@iklink You can’t. The software requires a CUDA enabled GPU.
as of 2021 Zephyr is a 14 day trial then you need to pay to keep using it
Good to know. Maybe just go with Meshroom.