Same Image, Different Score?
Auto-discovered
Visit resource ↗Description
Same Image, Different Score? is a February 2026 Halide Compression blog post by the Halide Team on how chroma handling in decode and post-processing changes image-quality metric scores for the same encoded bitstream.
It discusses YCbCr 4:2:0 versus 4:4:4, WebP's mandatory 4:2:0 limit, Sharp YUV preprocessing versus encoder efficiency, and decoder choices across FFmpeg, jpegli, ImageMagick, dwebp fancy upsampling, and dav1d wrappers. Experiments on the gb82 set compare jpegli, libwebp, Iris-WebP, and SVT-AV1 with metrics such as fssimu2, Butteraugli, and fcvvdp, showing multi-percent score swings from chroma scaling alone.
It is for intermediate to advanced compression engineers and platform teams who benchmark encoders fairly. Readers should know basic chroma subsampling and perceptual metric workflows.
The post is free HTML with charts, interactive perceptual comparisons, and a pointer to the open-source decbench scripts used for the runs.
It is useful when interpreting codec bake-offs so decoder and upsampling choices are not mistaken for pure encoder gains.