Pure Pytorch implementation of RGB to HSV/HSL conversion

This repository implement pytorch Tensor color space conversion from rgb to hsv/hsl and backwards.

Notes

1. The conversion process is differentiable in a natural way, but since the mapping from RGB to HSV/HSL space is not continuous in some place, it may not be a good idea to perform back propagation.

2. The code was written to accept only image like tensors tensor.shape == [B x 3 x H x W], but it's easy to modify the code to accept other shapes.

3. Reference is in here

Usage

example usage can be find in test.py. Before using the function, please make sure that:

1. The input/output rbg/hsv/hsl tensors should all be normalized to 0~1 for each channel
2. The rgb format is RGB instead of BGR
3. The shape of the tensor match the requirement