Explore alternative color quantization algorithms #2681
Description
Please describe the problem you're trying to solve
Sixel is an ancient image format that only supports 256 colors. So, when rendering a Sixel image, it is necessary to extract the 256 most representative colors from the image to serve as a palette and map any color in the image to one of these.
Yazi initially selected the neural-network quantization algorithm because it is, to my knowledge, the best in terms of quality among the available quantization algorithms. However, compared to other quantization algorithms, it requires a higher computational load, which is not ideal for low-spec devices.
It would be great to find an algorithm that delivers quality comparable to the neural-network quantization algorithm but is much faster.
Currently, Yazi provides the sixel_fraction configuration for the neural-network quantization algorithm, allowing users to balance between quantization precision and computational cost — that is, users can reduce CPU usage by increasing the sixel_fraction. So, when evaluating other quantization algorithms:
- The algorithm's quality should be comparable to that of
sixel_fraction = 10(highest quality). - The algorithm's speed should be considerably faster than
sixel_fraction = 30(highest speed).
Would you be willing to contribute this feature?
- Yes, I'll give it a shot
Additional context
No response
Checklist
- I have searched the existing issues/discussions
- The latest nightly build doesn't already have this feature