cjxl[v0.8.2] quality option not work about libjxl HOT 7 CLOSED

@DejayRezme The problem comes from the fact that for jpeg there is a difference between just storing the pixels losslessly, i.e. distance = 0 and losslessly compressing the bytes of the jpeg file, so that the original jpeg file can be reconstructed. So we need a way to specify a difference between
cjxl -d 0 --lossless_jpeg 1 foo.jpg foo.jxl
and
cjxl -d 0 --lossless_jpeg 0 foo.jpg foo.jxl
(which is probably not what you want)
and
cjxl -d 1 --lossless_jpeg 1 foo.jpg foo.jxl

We tried to make the defaults such that for jpegs lossless recompression of the bytes is the default, because this way no additional artifacts will we introduced, because the operation is reversible.
Do you have concrete suggestions what to change about these defaults?

from libjxl.

Thanks for the report!

This is intended behavior. In case you are compressing a jpg file, cjxl will by default attempt to compress it as lossless jpeg re-compression and passing a quality parameter will give the error message you are seeing.

It should encode the image with lossy compression when you pass the --lossless_jpeg=0 flag. The default for this flag is "1", since you didn't pass a value for this flag it will be equivalent to passing --lossless_jpeg=1.
I think we could make a better job explaining this in cjxl -h, currently we don't even mention the default setting for this flag

from libjxl.

thanks @mo271 , is --lossless_jpeg a hidden option, I can't find it when run -h

cjxl -h
JPEG XL encoder v0.8.2 [AVX2,SSE4,SSSE3,SSE2]
Usage: cjxl INPUT OUTPUT [OPTIONS...]
 INPUT
    the input can be PNG, APNG, GIF, JPEG, EXR, PPM, PFM, or PGX
 OUTPUT
    the compressed JXL output file
 -d maxError, --distance=maxError
    Max. butteraugli distance, lower = higher quality.
    0.0 = mathematically lossless. Default for already-lossy input (JPEG/GIF).
    1.0 = visually lossless. Default for other input.
    Recommended range: 0.5 .. 3.0. Allowed range: 0.0 ... 25.0.
    Mutually exclusive with --quality.
 -q QUALITY, --quality=QUALITY
    Quality setting (is remapped to --distance).    100 = mathematically lossless. Default for already-lossy input (JPEG/GIF).
    Other input gets encoded as per --distance default,
    which corresponds to quality 90.
    Quality values roughly match libjpeg quality.
    Recommended range: 68 .. 96. Allowed range: 0 .. 100.
    Mutually exclusive with --distance.
 -e EFFORT, --effort=EFFORT
    Encoder effort setting. Range: 1 .. 9.
     Default: 7. Higher number is more effort (slower).
 --brotli_effort=B_EFFORT
    Brotli effort setting. Range: 0 .. 11.
    Default: 9. Higher number is more effort (slower).
 -p, --progressive
    Enable progressive/responsive decoding.
 --resampling=-1|1|2|4|8
    Resampling for extra channels. Default of -1 applies resampling only for low quality. Value 1 does no downsampling (1x1), 2 does 2x2 downsampling, 4 is for 4x4 downsampling, and 8 for 8x8 downsampling.
 -v, --verbose
    Verbose output; can be repeated, also applies to help (!).
 -h, --help
    Prints this help message (use -v to see more options).

from libjxl.

thanks @mo271 , is --lossless_jpeg a hidden option, I can't find it when run -h

Yeah, that as well, it will show up only when running cjxl -h -v ( with at least one -v). Might it make sense to either show this option already without the -v or put it in the INPUT section like below? What do you think, @jonsneyers ?

JPEG XL encoder v0.8.2 [AVX2,SSE4,SSSE3,SSE2]
Usage: cjxl INPUT OUTPUT [OPTIONS...]
 INPUT
    the input can be PNG, APNG, GIF, JPEG, EXR, PPM, PFM, or PGX.
    If the input is JPEG, then by default lossless recompression will be applied.
 OUTPUT
    the compressed JXL output file

from libjxl.

Copy it, thank you so much.

from libjxl.

This broke my ps script for reencoding jpgs.

It seems now every user has to use this more complicated usage and add an if/then/else case into calling cjxl.

I do not see the point as to why - If I specify a distance, it's obvious I do not want lossless. If I specify distance = 0, it's obvious I want lossless. A tool should be as easy and comfortable to use as possible.

Why change the usage suddenly so that other tools or scripts using it break?

from libjxl.

Thanks for the reply @mo271

I would suggest that --lossless_jpeg is ignored if distance isn't 0. That someone needs --distance=0 and --lossless_jpeg=0 seems very rare, since the size should always certainly go up.

Or said differently, the default for --lossless_jpeg is 1 if distance=0,
and --lossless_jpeg is 0 (or ignored) if distance > 0

from libjxl.