WebP support
when saving images with 3 channels, tensorboard uses JPEG format. when saving images with 4 channels, tensorboard uses PNG format, which is not compressed.
WebP gives better compression ratios than any of the currently supported image formats.
How much work would it take to update this? What are the risks, if any?
Potential risk is that some very old browsers may not support webp. considering the disk and traffic saved and huge speed boost for loading 4 channels images, this will definitely worth it.
This is unlikely to receive priority development within Google, but contributions are always welcome. Even taking the time to go through which libraries would need to be added (if any) and pointers to which code points require change would be helpful.
@bileschi CLA keeps failing for me for some reason. but it's a one line change anyways. feel free include it in your commit
i can write webp from pytorch, and the display will be handeled by the browser.