Own T.81 encoder: Annex K tables, JFIF YCbCr 4:4:4/4:2:0 + gray, EXIF/XMP/ICC segments, bands of MCU
rows separated by RSTn so entropy coding runs on all cores. Per-core speed matches jpeg-encoder at 4:4:4
(225 vs 237 ms on 24 MP, single thread); with all cores 24 MP encodes in ~130 ms vs ~500–800 ms
(loaded machine). Output decodes with jpeg-decoder and zune-jpeg (odd sizes tested); same-or-smaller
files at equal/higher PSNR. 4:2:2 keeps jpeg-encoder. Bench: crates/codecs/examples/jpeg_bench.rs.
Co-Authored-By: Claude Opus 5.5 <[email protected]>