Contrast weighted SSIM + OKLAB colourspace for human perception

I applaud you on being able to get such stellar results without adversarial loss!

Would you be able to add contrast weighted SSIM to your loss function and see if gives a significant boost during training? It did in the halftoning task, a task you might think would be exhausted in optimisation after so many years!

[Efficient Halftoning via Deep Reinforcement Learning](https://arxiv.org/abs/2304.12152)

I also liked what you said about PSNR. Another low hanging fruit that I haven't seen any project take up yet is switching the images to a perceptually uniform colour space before AE/VAE training. Basically, using RGB, HSV, LAB etc uses the "wrong" colour difference formula for human perception. It's like every paper uses a PSNR measure that is the wrong "measuring stick", because everyone else uses the wrong measuring stick!

[A perceptual colour space for image processing](https://bottosson.github.io/posts/oklab/)