Cool-chic
Image coding performance¶
We provide comprehensive experimental results demonstrating the compression performance of Cool-chic, its low decoding complexity and the associated encoding complexity.
We also provide some already encoded bitstreams in samples/bitstream/ so
that everybody can decode them to try out our fast CPU-only neural decoder.
Image compression performance are presented on the kodak (RGB), clic20-pro-valid (RGB) and jvet (YUV420) datasets. The overall MSE is obtained by averaging the 3 channels weighted by their respective resolution. More information about obtaining one single MSE for the entire dataset is discussed in this github issue.
BD-rate¶
The following table sums up the compression performance obtained by Cool-chic 5.0 against different anchors. Results are measure through BD-rates which represents the relative rate required for Cool-chic to obtain the same quality as some other codec. Cool-chic 5.0 decoding complexity is set to 2 000 MAC / pixel for all datasets (HOP configuration).
| BD-rate of Cool-chic 5.0.1 vs. [%] | ||||||
|---|---|---|---|---|---|---|
| Cheng | ELIC | MLIC++ | Cool-chic 4.2 | Cool-chic 5.0.0 | VVC (VTM 28.3) | |
| kodak (RGB) | -10.4 % | +0.1 % | +10.2% | -6.1 % | -1.2 % | -3.9 % |
| clic20-pro-valid (RGB) | -21.3 % | -9.5 % | +0.5 % | -9.9 % | -0.6 % | -11.6% |
| jvet A (YUV420) | / | / | / | / | -1.4 % | -4.0 % |
| jvet B (YUV420) | / | / | / | -13.7% | -1.9 % | +6.6 % |
Rate-distortion graphs¶
CLIC20 Pro Valid¶
Kodak¶
