Open Enet4 opened 1 month ago
Hi, apologies for the very very late reply. I believe this is in tandem with the zune-jpegxl encoder
The jpeg-xl encoder is based on https://github.com/libjxl/libjxl/blob/c6355600c5932e3a9f1b95a4cadfa9b718fbec8f/lib/jxl/enc_fast_lossless.cc (libjxl fast lossless encoder) with the SIMD part stripped out, so the effort is directly tied to that, effort here means that we increase grid search space for sample collection.
To answer your question though
Is there a universally accepted range of values for effort? Should we also provide a number in the range 0..100?
No there should be though
If an encoder supports mathematically lossless encoding, does quality = 100 imply lossless encoding? Or is there another way to signal that the encoding is intended to be lossless?
I haven't gotten to the part where I have both a lossless and a lossy encoder for one format (that would be either webp or jxl), but it is usually better to have an explicit switch for such things, quality =100 is not equal to lossless when one chooses a lossy encoder, so separate them
Does quality still affect the encoding effort when doing lossless encoding, or is this part of the documentation outdated?
No it should not, this is used for the lossy encoders, specifically jpeg
Thanks for the reply. It was not originally clear that zune-jpegxl
only supported lossless encoding exclusively (it was clear that it offered lossless encoding, but not that it did not provide other forms of encoding, but maybe I extrapolated a bit too hard).
Is there a universally accepted range of values for effort? Should we also provide a number in the range 0..100?
No there should be though
If an encoder supports mathematically lossless encoding, does quality = 100 imply lossless encoding? Or is there another way to signal that the encoding is intended to be lossless?
I haven't gotten to the part where I have both a lossless and a lossy encoder for one format (that would be either webp or jxl), but it is usually better to have an explicit switch for such things, quality =100 is not equal to lossless when one chooses a lossy encoder, so separate them
Is there something the community can do to assist in the process of uniformizing the quality/lossless/effort parameters?
While tinkering with the encoding APIs in zune-image, I found an oddity in the type
EncoderOptions
. Documentation for the getter and setter ofeffort
is missing, and there are claims in methodget_quality
that thequality
option also affects effort when doing lossless encoding:It would be great if the documentation of the encoder options answered these questions:
effort
? Should we also provide a number in the range0..100
?quality = 100
imply lossless encoding? Or is there another way to signal that the encoding is intended to be lossless?quality
still affect the encoding effort when doing lossless encoding, or is this part of the documentation outdated?