Quoting Pierre-Anthony Lemieux (2024-11-18 06:57:52) > On Mon, Nov 11, 2024 at 12:19 AM WATANABE Osamu > <owata...@es.takushoku-u.ac.jp> wrote: > > > > I have confirmed that these failures in FATE were due to the insufficient > > floating point precision of a 32-bit environment. > > > > The commit 82467b635efced67c1767cb810af1f3c31a2e493 introduces the improved > > dequantization in FF_DWT97_INT path and makes FFmpeg compliant for ISO/IEC > > 15444-4 (Conformance testing.) > > > > Actually, both 64 and 32-bit FFmpeg with the commit pass the conformance > > testing. > > > > Current FATE tests for JPEG 2000 are to check the CRC of reconstructed > > images. > > Checking CRC is unsuitable for lossy codestream because even a patch that > > can improve the quality of a reconstructed image cannot pass the FATE. > > Passing FATE is, of course, important, but passing the official (ISO's) > > conformance tests is essential, too. > > > > I recommend introducing a way to allow some tolerance in FATE tests for > > JPEG 2000 lossy codestreams. > > For example, including the original files and using the PSNR value in FATE > > will be an option. > > The original files are approximately 10 MB. Any objection/concerns to > adding them to FATE, which can then apply the conformance requirements > specified in the standard?
IMO it should not be a problem, other codecs have far larger conformance samples. -- Anton Khirnov _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
