Cool-chic (pronounced /kul ʃik/ as in French 🥖🧀🍷) is a low-complexity neural image codec based on overfitting. It offers image coding performance competitive with H.266/VVC for 1000 multiplications per decoded pixel.
- New and improved latent upsampling module
- Leverage symmetric and separable convolution kernels to reduce complexity & parameters count
- Learn two filters per upsampling step instead of one for all upsampling steps
- 1% to 5% rate reduction for the same image quality
- 30% complexity reduction using a smaller Auto-Regressive Module
- From 2000 MAC / decoded pixel to 1300 MAC / decoded pixel
- 10% faster decoding speed
Check-out the release history to see previous versions of Cool-chic.
More details are available on the Cool-chic page
# We need to get these packages to compile the C API and bind it to python.
sudo add-apt-repository -y ppa:deadsnakes/ppa && sudo apt update
sudo apt install -y build-essential python3.10-dev pip
git clone https://github.com/Orange-OpenSource/Cool-Chic.git && cd Cool-Chic
# Install create and activate virtual env
python3.10 -m pip install virtualenv
python3.10 -m virtualenv venv && source venv/bin/activate
# Install Cool-chic
pip install -e .
# Sanity check
python -m test.sanity_checkYou're good to go!
The Cool-chic page provides comprehensive rate-distortion results and compressed bitstreams allowing
to reproduce the results inside the results/ directory.
| BD-rate of Cool-chic 3.4 vs. [%] | Avg. decoder complexity | |||||||
|---|---|---|---|---|---|---|---|---|
| Cheng | ELIC | Cool-chic 3.3 | C3 | HEVC (HM 16) | VVC (VTM 19) | MAC / pixel | CPU Time [ms] | |
| kodak | -4.2 % | +7.5 % | -0.9 % | -4.3 % | -17.2 % | +3.4 % | 1303 | 74 |
| clic20-pro-valid | -13.2 % | -0.2 % | -0.3 % | -1.3 % | -25.1 % | -2.3 % |
1357 | 354 |
| jvet | / | / | -0.2 % | / | -18.3 % | +18.6 % | 1249 | 143 |
Decoding time are obtained on a single CPU core of an an AMD EPYC 7282 16-Core Processor
PSNR is computed in the RGB domain for kodak and CLIC20, in the YUV420 domain for jvet
Special thanks go to Hyunjik Kim, Matthias Bauer, Lucas Theis, Jonathan Richard Schwarz and Emilien Dupont for their great work enhancing Cool-chic: C3: High-performance and low-complexity neural compression from a single image or video, Kim et al.
