Open leo-barnes opened 3 years ago
Maybe a good first step would be to provide the APIs to take an existing AVIF image and make it the thumbnail of another (maybe those already exist). We could integrate that into Squoosh.app.
Maybe a good first step would be to provide the APIs to take an existing AVIF image and make it the thumbnail of another (maybe those already exist).
I think GPAC/MP4Box might support doing that already. Not extremely familiar with it though. @tdaede probably knows.
I'd like blurring to be automatically handled in pre-processing (inside image), or in post-processing (by browser). CSS blurring is bad for performance when images are displayed into an infinite virtual scroller, and the user scrolls fast.
With talks of blurring... can we make sure whatever blurring route is taken it's a linear gamma blur and not an sRGB blur (i.e. CSS's filter: blur(16px)
) like some of the demos in the previous thread? The blobs of muddy dark colors look bad and aren't "correct".
I only used CSS's blur because it was there. A real implementation needs to be a lot faster, and as you say, linear.
So what happened to this?
From the discussion in this av1-avif issue and the last SG-Image meeting, the conclusion was that the spec now has everything needed for doing progressive refinement either via a separate thumbnail image or via multi-layer coding.
It was concluded in the meeting that the next step would be for implementations and tooling to start using it. Adding a downscaled, very low quality thumbnail to images would be a good step as shown in @jakearchibald's experiments in the issue linked above. @skal65535 mentioned that avifenc currently does not support adding a thumbnail, or at least not to add an automatically downscaled thumb with low quality. Adding that functionality would be a good way for people to start generating content that can then be used to test decode implementations.