This doesn't seem to be gamma correct, I just see / 255 and * 255.
sRGB is a nonlinear colour space and so you can't do linear operations in that space (because a^2 + b^2 isn't (a+b)^2 in general).
Show HN: High End Color Quantizer
21–30 of 43 posts
Re: Show HN: High End Color Quantizer
#22What's the primary use case you had in mind here? In the example I see it generating a palette of 256 colors and then using them, but it doesn't seem to correspond to any modern use case. AFAIU one currently needs dithering either as part of print/display process (but then you have a fixed palette), or for compression, but I think this makes sense nowadays only with very low color count, like 16 max?
Also, dithering and color quantization are two vastly different operations on two different data types in two different domains that don't belong in the same topic at all.
Still, color quantization is a really interesting rabbit hole to go down if you're new to graphics programming, or at least it was for me. It's a mixed blessing that almost nobody has to confront the problem anymore.
Re: Show HN: High End Color Quantizer
#23What's the primary use case you had in mind here? In the example I see it generating a palette of 256 colors and then using them, but it doesn't seem to correspond to any modern use case. AFAIU one currently needs dithering either as part of print/display process (but then you have a fixed palette), or for compression, but I think this makes sense nowadays only with very low color count, like 16 max?
I'm always interested in ways to increase the quality of GIF rendering. There are absolutely tons of places that still need GIF support, either because they don't allow video uploads, or because the videos don't auto-play.
Gifski uses the png-quant library, and I wonder how this compares?
Re: Show HN: High End Color Quantizer
#24What is color quantization used for in this lossy compression age?
GIFs
Re: Show HN: High End Color Quantizer
#25Can this also do quantizations like median cut?
Does it also support colour spaces other than RGB, CIEDE (I think I saw that in the source), i.e CMYK for paint mixing, and similar ilk?
I have a few personal projects that would benefit from a library that is wide ranging in the colour space, dither algorithms (I only saw riezsma) and quantizations.
Typically these are all usually implemented in different libraries :(
Thanks!
Re: Show HN: High End Color Quantizer
#26This doesn't seem to be gamma correct, I just see / 255 and * 255. sRGB is a nonlinear colour space and so you can't do linear operations in that space (because a^2 + b^2 isn't (a+b)^2 in general).
Gamma aware operations happen in the C code.
The python code you're referencing is just changing the scale of color intensities. What you shouldn't do is liberally add up sRGB colors, take averages and generally do any math on them unless you're aware of the non-linearity of the space.
Re: Show HN: High End Color Quantizer
#27> eaaaarly stage […] beta
If your goal is for users to adopt the use of the software, then you can easily increase acceptance by going the proverbial extra mile:
1. Make it installable via `uv tool install patolette` with the optimisations taken care of automatically. 2. Compare its results in the documentation/on the project Web site against the incumbents. https://news.ycombinator.com/item?id=26646035 Find standard test images https://ddg.gg/?q=color+quantization+test+corpus , copy the split-image/slider technique from https://uprootlabs.github.io/poly-flif/
The rationale for this is that each interested user should not have to replicate this work on his own.
Re: Show HN: High End Color Quantizer
#28What specifically about this paper caught your eye that you wanted to implement that, what does it do better than other methods? Can you give a quick primer on what it does, and what the optional kmeans refinement does?
Something that caught my eye is that it seemed to be a kind of "controlled" K-Means. One problem with K-means is that it's too sensitive to the initial state. You can run it multiple times with different initial states or use fancy initialization techniques (or both) but even then nothing really guarantees you won't be stuck with a bad local optimum. Another thing was that the guy that wrote the paper also authored an insanely high quality method the year before and claimed this one was better. Not seeing any available implementations I wondered how good it actually was.
The optional K-Means step just grabs whatever palette the original method yielded and uses it as initial state for a final refinement step. This gives you (or gets you closer) to a local optimum. In a lot of cases it makes little difference, but it can bump up quality sometimes.
Re: Show HN: High End Color Quantizer
#29What's the primary use case you had in mind here? In the example I see it generating a palette of 256 colors and then using them, but it doesn't seem to correspond to any modern use case. AFAIU one currently needs dithering either as part of print/display process (but then you have a fixed palette), or for compression, but I think this makes sense nowadays only with very low color count, like 16 max?
Also, dithering and color quantization are two vastly different operations on two different data types in two different domains that don't belong in the same topic at all. Still, color quantization is a really interesting rabbit hole to go down if you're new to graphics programming, or at least it was for me. It's a mixed blessing that almost nobody has to confront the problem anymore.
Really? Dithering is generally only useful with quantized colors, you can't dither something that's already quantized without knowledge of the original, and many/most people who want to do quantization also want to do dithering. The algorithms themselves might not be conceptually similar, but for practical purposes they seems very related.
Re: Show HN: High End Color Quantizer
#30The example image only shows the differences between parts of the image that are deemed as salient, but does not show the effects of the tradeoff on the non-salient parts nor does it show an entire quantized image. I'd say that showing full example results are the most important part of showcasing a "high end" color quantizer.
Yeah, good point. I'll definitely add an image showing the saliency map tradeoff.
Regarding full examples, because some other projects seem to have cherry picked cases where they perform very well, I wanted to go for a "try it out yourself" approach, at least for now. Maybe in the future I'll add a proper showcase. Thanks for the feedback :)