I just wanted to drop by and express my heartfelt appreciation for the amazing content you've been sharing on your channel. Your videos are not only educational, but also a huge source of inspiration for me and many others in the coding community. The generosity with which you share your knowledge and passion for coding is truly commendable. Thank you for being such a fantastic resource and for all the hard work you put into making complex concepts accessible and fun. Keep up the incredible work!
Would like to second this as well Daniel for making both education and programming fun and at the same time. I've watched you for years and years and I have always enjoyed everything that you've made, especially the coding challenges. For a little while, I even replicated some of the coding challenges in Python using the p5 package for python. Keep up the great stuff and don't ever stop Sir!
In college a was an undergraduate research assistant converting a 2d voronoi lattice simulation to 3d. 7 years later I think I finally have a grasp on what that simulation was actually meant to do because of this video so thank you.
I love your enthusiasm and I really appreciate that you show your mistakes and laugh about them. Let's be honest, learning to program is boring. You make it really fun.
The Coding Train is by far the most awesome educational and inspiring coding lessons there is. Thank you so much for sharing your coding adventures, you are the best. ♥
Wow I admire your coding skills. I could not do what you keep doing. I don't have the passion to always throw myself into new coding problems. I will never forget you, "Coding Rainbow" 🙏
In all my (few) but intense years of learning creative coding I have not had a greater source of learning and inspiration from your channel. Your ways of teaching and exploring deep and complex topics of creative coding are amazing, fun, and highly enjoyable. The coding community cannot thank you enough!
Dude, you are doing a great job. You are personally entertaining, you are providing problem solving techniques and the projects you are doing are very cool. I'm glad I found this channel.
Wonderful video, Daniel. I love how each video is so visual and leads me to learn a new algorithm or mathematical technique as well as improving my coding.
This is basically just shader math, you should try doing a video about/using glsl or hlsl, theres some fun stuff you can do with them, maybe use shader toy? (Never used it bc Im a gamedev so use shaders in engine/in opengl)
Every time you introduce some idea, I feel like, "yeah sure, I know the math, that's how that works, no way you can implement it in a single video", then you simply code it up. I am amazed.
I felt so relaxed watching this 😌with a nice touch of Dan's enthusiastic sounds every time he got the next thing to work. Jokes aside! Got lots of inspiration from this tutorial - thank you Dan!
This algorithm is good for relaxing UV maps for unfolded meshes for artist texturing. The "raster dots" effect can easily be achieved by averaging the image into boxes of n by n pixels and drawing a circle scaled to the magnitude of the average value. Instead of scaled the dot can also be proportionally occluded by another offset dot masking it out, or a hole dot within the dot and whatnot...
Love it! It reminds me of a technique you used before, which is Floyd-Steinberg dithering. However, this one seems more challenging since you used an external library, but it sets a challenge to write the whole algorithm from scratch and optimize it, which I think I will attempt!
Floyd Steinberg performs error diffusion in some direction, whatever direction you're scanning towards, say towards bottom right, so if you apply it to a moving image, your start of scan is relatively stable say top left corner but your bottom right is very noisy. This isn't visible in static images, only moving. You could potentially start scanning out from the middle but that doesn't really solve the problem, it will still be obvious where the noise is eminating from. There's a more modern algorithm with a similar visual outcome to Floyd Steinberg but without bias in any direction, where you use a precomputed blue noise as a threshold function, and you can reuse the same across frames or use a spatiotemporal variant, and the way to compute this blue noise, the foundation is in this very video - Lloyd's Relaxation of a random point set. Blue noise sampling is the foundation of a lot of modern stochastic rendering methods in graphics. There's a whole research group at Nvidia concerned with blue noise.
You don't need voronoi for how it was eventually done. You could just determine the point that is closest to the pixel ... (That point's voronoi cell will contain the pixel.)
Hay i been braking my head on this algo for the last yr, was exploring how to go from delaney to nanite 😂 and went as mad as you have, 😅 the fact that u had a hard time, makes me not feel so dumb anymore, not to take anything away from you, love your stuff watch as many as i can, would love to see you experiment with shadertoy constraints and do some crazy stuff 😊
Just wanted to note that there are more sophisticated stippling algorithms, like the one based on power diagrams, which is a generalization of Voronoi diagrams (implementing it _is_ a coding challenge). See "Blue noise through optimal transport" paper for details. Besides, you check whether the center of each pixel is inside the Voronoi cell to calculate its mass center. Considering each pixel as a square and calculating the intersection area with each Voronoi cell will give a significant quality improvement for the Lloyd algorithm.
Back in the day I made a random map generator and the coastlines and elevation features were drawn via voronoi diagrams. I coded the main app in p5 but also imported d3 for the voronoi computations, so, yeah. No shame in that, lol.
fun fact: we just had those in our first lecture of solid state physics this monday. i'm deffinitely tempted to make a 3D version of this, and make it efficiently, that seems like a fun geometric problem.
Instead of converting points to voronoi, and then figuring out which polygon a pixel is in, you could just determine which point is closest to the pixel... (The trick with remembering the previous pixel's index still works, or you could use a k-d Tree to make it efficient.)
Maybe I should leave a comment on each video you upload. Whenever I'm coding or facing a task, I play one of your challenge videos. They inspire and empower me to overcome the obstacles in my tasks.
@4:31 Other ways to convert an array of x/y into a flat array of pairwise numbers: const somePoints = [{x:1,y:2},{x:3,y:4},{x:5,y:6}]; const method2 = []; somePoints.forEach(p => { method2.push(p.x); method2.push(p.y); }); console.log(JSON.stringify(method2)); const method3 = somePoints.reduce((vec, p) => vec.concat([p.x, p.y]), []); console.log(JSON.stringify(method3)); .forEach is quite a bit better than writing for loops. .reduce is for functional programming weenies, although it is admittedly cool and tends to be compact.
It’s been YEARS (about 5 to 8 years) that i was looking something exactly that! It would have been immensely interesting to adapt it as a lizard skin! I’m now curious how it will be done with blender (open source 3D modeling software) my idea was to use the white color as inexistant scale while the black color would represent the smallest scale of a lizard. Your example is exactly what I’m trying to reproduce as a lizard skin generator! Anyway, thank again for the amazing video you did here!
0:39 Yeah! I kind of am a little confused about that particular thing, currently what I've always thought them to be was when you take a bunch of points with a given radius, then starting with the highest radius and going down just render every circle, then bits closer to a point will be prioritized over ones further resulting in segments being created. But it was never really clear whether this was actually Voronoi or just another simpler method intended to replicate it.
8:29 Ah! So, it sounds like the method I learned actually is Voronoi and it's much more efficient for finding the areas than the method of having each pixel calculate their distance.
Man.. I really love your videos although I don't code almost at all.. but damn i like how you integrate things that are pure abstractions for me into applicable real projects. You are a true professor.
Small mistake on 9:07, where the last circumcircle of the right triangle should be. But overall, that was a fun and educational coding challenge! As always ;)
While unrelated to this particular video, surely this comment section is as good as any! I was thinking about how it may be nice to see what you can do with exclusively bill-boarding in 3d graphics, there's of course traditional bill boarding, but you can also make other bill boarding rendering techniques, like for instance rendering 3d lines, this is where you position your image in-between the projected versions of two points then scale it up according to the distance between them and rotate it to fit between, this works really well for a similar cost to standard bill-boarding, you can also do something similar to traditional bill boarding but where you place the image at the incenter of the projected version of a 3d triangle and of course scale it to the size of the incenter too, though I haven't worked out how you could also try incorporating rotation into that to make it a little better. Regardless, I'm curious just how far you could go with just methods like these, I've already made some interesting things with it like a spider since spiders are mostly just lines anyway that could be made out of the 3d line rendering, similarly skeletons would be easy, but for anything beyond that you may need to get super artistic with both traditional and incenter bill boarding plus possibly making new rendering techniques.
Hm, I think the fact that you've once again reverted back to averages instead of properly finding the centroid but weighted, the black hole problem is likely still back at least to an extent. If your image was either consistent brightness or a brightness according to the distance from some vector, then the points would likely all collapse into the same place.
Hm, after some thought, here's a potential way to weight the centroid towards certain values, first you compute the normal centroid, then you interpolate each Vertice from itself to the centroid based on its weighting then you recompute the centroid with the new vertices to get the weighted version. I don't know if this is the best method of doing it, but it's worth a shot!
15:08 it would be cool if you wrapped the averaging code in an "if then" with a size greater than or equal to like 9 pixels or something to limit how deep the blackhole became
I barely understood anything but all I understood was "little dots move around and make cool shapes, man codes them to make image"
9 месяцев назад
Would it be possible to use both diagrams and other algorithms shown in this video to let's say, triangulate a person's unknown position? I'm no spy nor stalker of any sort, just remembered I was addicted to tv series called Numbers which used Math to solve crimes (of course in Hollywood style, but you get the point). My idea here is how Math can be used to almost anything in life, rather than drawing (which is awesome too).
Who ever decided on "ducks" at Columbia Collage was a genius, you'd have to be to pick an animal that just says quack all day long as they walk up and down the sidewalk.
I feel like maybe you could do this with spatial hashing. Like an oct tree, for each point we get its weight, and we find the nearest seed point at each iteration or something? idk. Sounds like something that'd be quite fast on gpu.
I hope one of you train enthusiasts can answer this - how did you know the delaunay portion of the d3 library was available from the cdn (with that particular url)? I looked on the d3 site and I couldn't find the option of linking to just portions of the d3 libraries without using import statements etc. I much prefer your method of just putting it as a source in index.html, but don't know how to find that it exists? Great video!
If you created a circle on every point which all expand at the same rate and had them ‘flatten’ when they collided with another circle would that create the voronoi diagram? I’m not sure how efficient that would be code-wise but I’m just curious as to whether it would work at all.
I love how this feels both like a University Lecture, and a kids educational channel!
the pinnacle of eli5
perfect videos for adults who loved bill nye as kids
I just wanted to drop by and express my heartfelt appreciation for the amazing content you've been sharing on your channel. Your videos are not only educational, but also a huge source of inspiration for me and many others in the coding community. The generosity with which you share your knowledge and passion for coding is truly commendable. Thank you for being such a fantastic resource and for all the hard work you put into making complex concepts accessible and fun. Keep up the incredible work!
100%!
Would like to second this as well Daniel for making both education and programming fun and at the same time. I've watched you for years and years and I have always enjoyed everything that you've made, especially the coding challenges. For a little while, I even replicated some of the coding challenges in Python using the p5 package for python. Keep up the great stuff and don't ever stop Sir!
I was going to say "This rules. You rule". But your's is more good words.
I'd love to see this done for each R G B channel individually. i think that would create a really cool comic book type of effect.
14:40 "This is lovely, but it's collapsing into an black hole..." That's a very funny quote 😂
it looks so organic there
Accidentally made a gravity simulator…
In college a was an undergraduate research assistant converting a 2d voronoi lattice simulation to 3d. 7 years later I think I finally have a grasp on what that simulation was actually meant to do because of this video so thank you.
how fascinating it was to see Gloria Pickle turning into Gloria pixels! you make it look so easy!
I program for a living and I couldn't agree more. He makes it looks so easy! But I know it's much more complicated, he's just so good at his job.
I love your enthusiasm and I really appreciate that you show your mistakes and laugh about them. Let's be honest, learning to program is boring. You make it really fun.
The Coding Train is by far the most awesome educational and inspiring coding lessons there is. Thank you so much for sharing your coding adventures, you are the best. ♥
This channel is the best programming channel on RUclips... You make the most difficult concepts simple.. amazing
Can’t believe this guy taught me so many wonders of coding back in the day and he’s still going! Keep up the good work!
You're doing a fantastic job. Don't ever stop creating!
As a former graphics programmer, I loved this! Also, your enthusiasm and positivity is infectious! Love it man, always happy to hop on board!
I love the new editing style Dan! Feels refreshing to see intros and the Apple ][ jingles being used in the beginning of the video!
This guy makes me so happy and is such an inspiration for RUclips and coding in general. I love it
Wow I admire your coding skills. I could not do what you keep doing. I don't have the passion to always throw myself into new coding problems.
I will never forget you, "Coding Rainbow" 🙏
In all my (few) but intense years of learning creative coding I have not had a greater source of learning and inspiration from your channel. Your ways of teaching and exploring deep and complex topics of creative coding are amazing, fun, and highly enjoyable. The coding community cannot thank you enough!
Dude, you are doing a great job. You are personally entertaining, you are providing problem solving techniques and the projects you are doing are very cool. I'm glad I found this channel.
Wonderful video, Daniel. I love how each video is so visual and leads me to learn a new algorithm or mathematical technique as well as improving my coding.
Mindblowing His ability to pick some abstract concept and just use it to solve an real problem in minutes
Watching these for free, what a great world your building, thank you!
This is basically just shader math, you should try doing a video about/using glsl or hlsl, theres some fun stuff you can do with them, maybe use shader toy? (Never used it bc Im a gamedev so use shaders in engine/in opengl)
Every time you introduce some idea, I feel like, "yeah sure, I know the math, that's how that works, no way you can implement it in a single video", then you simply code it up. I am amazed.
I felt so relaxed watching this 😌with a nice touch of Dan's enthusiastic sounds every time he got the next thing to work. Jokes aside! Got lots of inspiration from this tutorial - thank you Dan!
Thank you Patt!
Hands down....Black belt in coding. I love it.
This algorithm is good for relaxing UV maps for unfolded meshes for artist texturing. The "raster dots" effect can easily be achieved by averaging the image into boxes of n by n pixels and drawing a circle scaled to the magnitude of the average value. Instead of scaled the dot can also be proportionally occluded by another offset dot masking it out, or a hole dot within the dot and whatnot...
Love it! It reminds me of a technique you used before, which is Floyd-Steinberg dithering. However, this one seems more challenging since you used an external library, but it sets a challenge to write the whole algorithm from scratch and optimize it, which I think I will attempt!
oh, please do! I would love to see it!!!
Floyd Steinberg performs error diffusion in some direction, whatever direction you're scanning towards, say towards bottom right, so if you apply it to a moving image, your start of scan is relatively stable say top left corner but your bottom right is very noisy. This isn't visible in static images, only moving.
You could potentially start scanning out from the middle but that doesn't really solve the problem, it will still be obvious where the noise is eminating from.
There's a more modern algorithm with a similar visual outcome to Floyd Steinberg but without bias in any direction, where you use a precomputed blue noise as a threshold function, and you can reuse the same across frames or use a spatiotemporal variant, and the way to compute this blue noise, the foundation is in this very video - Lloyd's Relaxation of a random point set.
Blue noise sampling is the foundation of a lot of modern stochastic rendering methods in graphics. There's a whole research group at Nvidia concerned with blue noise.
You don't need voronoi for how it was eventually done.
You could just determine the point that is closest to the pixel ... (That point's voronoi cell will contain the pixel.)
@@landsgevaer the voronoi diagram is necessary for getting the polygons whose centroid the points move to.
@@a.lollipop Like I said, you don't need it anymore the way it was eventually done. I gave the much simpler alternative above...
Hay i been braking my head on this algo for the last yr, was exploring how to go from delaney to nanite 😂 and went as mad as you have, 😅 the fact that u had a hard time, makes me not feel so dumb anymore, not to take anything away from you, love your stuff watch as many as i can, would love to see you experiment with shadertoy constraints and do some crazy stuff 😊
Just wanted to note that there are more sophisticated stippling algorithms, like the one based on power diagrams, which is a generalization of Voronoi diagrams (implementing it _is_ a coding challenge). See "Blue noise through optimal transport" paper for details.
Besides, you check whether the center of each pixel is inside the Voronoi cell to calculate its mass center. Considering each pixel as a square and calculating the intersection area with each Voronoi cell will give a significant quality improvement for the Lloyd algorithm.
I understood the absolute number of ZERO things. Thank you.
The formula for the area of a polygon is the famous shoelace formula! Mathologer, among other people, has a nice video about it.
This has been my favorite channel since I don`t know, 2017
Back in the day I made a random map generator and the coastlines and elevation features were drawn via voronoi diagrams. I coded the main app in p5 but also imported d3 for the voronoi computations, so, yeah. No shame in that, lol.
My brain hurt, but I was amazed by the beauty of mathematics and your explanation. I have been following it more or less for years. thanks
I can't and have never coded in my life but i still love your videos!
You can. Just start somewhere and one step at a time.
Pretty good stuff. This seems a very useful method to correctly simulate film grain in a digital photo
fun fact: we just had those in our first lecture of solid state physics this monday. i'm deffinitely tempted to make a 3D version of this, and make it efficiently, that seems like a fun geometric problem.
babe wake up! new coding train video just dropped!
Thank you! Your work is inspiring 😄
The black hole! 😯 Unexpected but so cool!
This was super fun to watch, thanks so much!
Instead of converting points to voronoi, and then figuring out which polygon a pixel is in, you could just determine which point is closest to the pixel...
(The trick with remembering the previous pixel's index still works, or you could use a k-d Tree to make it efficient.)
Maybe I should leave a comment on each video you upload. Whenever I'm coding or facing a task, I play one of your challenge videos. They inspire and empower me to overcome the obstacles in my tasks.
You are brilliant...loving the videos, thanks 🙂
Im so excited! I saw your video on nebula tv but i didnt had any subscription so i couldnt watch i didnt think you would upload it this early
@4:31 Other ways to convert an array of x/y into a flat array of pairwise numbers:
const somePoints = [{x:1,y:2},{x:3,y:4},{x:5,y:6}];
const method2 = [];
somePoints.forEach(p => {
method2.push(p.x);
method2.push(p.y);
});
console.log(JSON.stringify(method2));
const method3 = somePoints.reduce((vec, p) => vec.concat([p.x, p.y]), []);
console.log(JSON.stringify(method3));
.forEach is quite a bit better than writing for loops. .reduce is for functional programming weenies, although it is admittedly cool and tends to be compact.
You are Bob Ross of programming
It’s been YEARS (about 5 to 8 years) that i was looking something exactly that! It would have been immensely interesting to adapt it as a lizard skin! I’m now curious how it will be done with blender (open source 3D modeling software) my idea was to use the white color as inexistant scale while the black color would represent the smallest scale of a lizard. Your example is exactly what I’m trying to reproduce as a lizard skin generator!
Anyway, thank again for the amazing video you did here!
well that was awesome
The black holee is mesmerizing
Thank you for another great video, Mr. Shiffman 😁👍
Such a great series!
the point where you made the points relax based on the centre of gravity could (maybe) be used for liquid simulation
14:43 God coding the fabric of spacetime
Hard work, and good explanation. Very helpful! 👍
I like dithering in art, it looks cool
amazing :D
0:39 Yeah! I kind of am a little confused about that particular thing, currently what I've always thought them to be was when you take a bunch of points with a given radius, then starting with the highest radius and going down just render every circle, then bits closer to a point will be prioritized over ones further resulting in segments being created. But it was never really clear whether this was actually Voronoi or just another simpler method intended to replicate it.
8:29 Ah! So, it sounds like the method I learned actually is Voronoi and it's much more efficient for finding the areas than the method of having each pixel calculate their distance.
I have no idea whats going on but this is so cool 😃
Man.. I really love your videos although I don't code almost at all.. but damn i like how you integrate things that are pure abstractions for me into applicable real projects. You are a true professor.
Small mistake on 9:07, where the last circumcircle of the right triangle should be. But overall, that was a fun and educational coding challenge! As always ;)
Well. This was pretty amazing
This does look like fun. I think I'll try it in C with RayLib, and use color. This might make for a really cool image viewer.
Coding Challenge 182: edge detection?
While unrelated to this particular video, surely this comment section is as good as any! I was thinking about how it may be nice to see what you can do with exclusively bill-boarding in 3d graphics, there's of course traditional bill boarding, but you can also make other bill boarding rendering techniques, like for instance rendering 3d lines, this is where you position your image in-between the projected versions of two points then scale it up according to the distance between them and rotate it to fit between, this works really well for a similar cost to standard bill-boarding, you can also do something similar to traditional bill boarding but where you place the image at the incenter of the projected version of a 3d triangle and of course scale it to the size of the incenter too, though I haven't worked out how you could also try incorporating rotation into that to make it a little better. Regardless, I'm curious just how far you could go with just methods like these, I've already made some interesting things with it like a spider since spiders are mostly just lines anyway that could be made out of the 3d line rendering, similarly skeletons would be easy, but for anything beyond that you may need to get super artistic with both traditional and incenter bill boarding plus possibly making new rendering techniques.
Stunning as always!
Thank you!
Could be a beautiful effect to transform from the totally distributed dots to those of the image.
He already did it with the logo at 0:09 and it looks so cool!
Okay but how? Simply the reversed process?
Yes I think so
It's a translation of light and shadow with a circle radius variation.
I love your content. If you allow me, here is a suggestion. You could implement an algorithm to find the path to the goal on the Micro Mouse Race.
Wow! Last week, I wrote a kinetic voronoi algorithm which doesn't require any triangularization
Cool as always
Hm, I think the fact that you've once again reverted back to averages instead of properly finding the centroid but weighted, the black hole problem is likely still back at least to an extent. If your image was either consistent brightness or a brightness according to the distance from some vector, then the points would likely all collapse into the same place.
Hm, after some thought, here's a potential way to weight the centroid towards certain values, first you compute the normal centroid, then you interpolate each Vertice from itself to the centroid based on its weighting then you recompute the centroid with the new vertices to get the weighted version. I don't know if this is the best method of doing it, but it's worth a shot!
Good one Dan !
“Let's put them into an array. And lerp the original points towards the centroid.” It seems weird that I understand that...
I'm currently learning Apps Script for work as a Python person in private, and this video gave me violent flashbacks.
This is actually part of Solid state Physics formulation of the K-Space for electrons
You are the best !
oh man ive been wanting to do this
I like ,relax, followed by head scratching 😂😂
Amazing video!
fascinating and relaxing
Amazing content, thanks !
Wow, so cool!!! ❤
15:08 it would be cool if you wrapped the averaging code in an "if then" with a size greater than or equal to like 9 pixels or something to limit how deep the blackhole became
I barely understood anything but all I understood was "little dots move around and make cool shapes, man codes them to make image"
Would it be possible to use both diagrams and other algorithms shown in this video to let's say, triangulate a person's unknown position? I'm no spy nor stalker of any sort, just remembered I was addicted to tv series called Numbers which used Math to solve crimes (of course in Hollywood style, but you get the point).
My idea here is how Math can be used to almost anything in life, rather than drawing (which is awesome too).
geez xD this is high vibration content
Loved it
Is it maths, cs or stand up comedy though
Who ever decided on "ducks" at Columbia Collage was a genius, you'd have to be to pick an animal that just says quack all day long as they walk up and down the sidewalk.
I feel like maybe you could do this with spatial hashing. Like an oct tree, for each point we get its weight, and we find the nearest seed point at each iteration or something? idk. Sounds like something that'd be quite fast on gpu.
This looks like a great way to create something that can be fed to a pen plotter. I imagine a giant stipple plot of a picture of my cat :D
I hope one of you train enthusiasts can answer this - how did you know the delaunay portion of the d3 library was available from the cdn (with that particular url)? I looked on the d3 site and I couldn't find the option of linking to just portions of the d3 libraries without using import statements etc. I much prefer your method of just putting it as a source in index.html, but don't know how to find that it exists? Great video!
Excellent!! :)
Awesome!
thats cool!
Thanks!
Reminded me of Origami and Crease Pattern
Lovely video but I lost count of the misspelled Dalaunays? 😅
I wonder if there are simpler ways to achieve the same effect, like have the points attracted to dark pixels and repeled by eachother.
I wonder, is this algorithm used in high-end RIPs to convert gray-scale image to BW one for printing?
If you created a circle on every point which all expand at the same rate and had them ‘flatten’ when they collided with another circle would that create the voronoi diagram? I’m not sure how efficient that would be code-wise but I’m just curious as to whether it would work at all.
The moment I finished reading the video's title is when I failed the challenge 😅