Valerii Zapodovnikov

BTW, >The A100 includes a 5-core hardware JPEG decode engine. nvJPEG takes advantage of the hardware backend for batched processing of JPEG images. So apparently A100 for servers can even...

In fact that FAQ is very strange and wrong because it can be NULL only if the allocation of memory happens (or if you somehow managed to assign NULL by...

Oh, you are the same guy. https://github.com/uclouvain/openjpeg/issues/589 BTW, here mode == kQuantModeRAW is shot curcuiting. So if it false...

> The only kind of YCbCr we support, is the one that is used in jpeg/jfif: full range YCbCr, And BT.601 matrix. BTW, Jpeg 2000 supports all kind of stuff:...

> Then again, you can probably construct an icc profile for all of those... No, those define "matrix". For example, PhotoYCC has different quantisation (integer to float), not simple 16-236,...

> If you want to use a different ycbcr matrix, you could define an icc profile that includes that matrix already, and encode the image like that. Matrix of ICC...

Interesting what -E 3 does and why it is effective. https://github.com/libjxl/libjxl/blob/2f872a190f79390ba27a66b2ecb14d51114c12ef/tools/cjxl.cc#L441-L443

> Looks like the decoder is too slow for realtime, Libgif? P.S. You can just attach jxl files in github by drag and drop. Just rename it to .jpeg. it...

There is a report that it happens with ffmpeg too, but ffmpeg does not use your native gif decoder... Strange, https://www.reddit.com/r/ffmpeg/comments/ukjii7/comment/i7qxaja/

So there is no way to see it, like in JPEG 2000, where one transform is lossless 5/3 and the other (9/7) is lossy... Well, in the end is not...