🎲 Cool-chic 4.2: Common Randomness & Wasserstein Distance! 🎲

This release focuses on perceptually-oriented image coding. This release draws heavily on the following paper: Good, Cheap, and Fast: Overfitted Image Compression with Wasserstein Distortion, Ballé et al..

• Wasserstein Distance as a distortion metric: use --tune=wasserstein
• Decoder-side common randomness for additional details in the decoded image
• Improved image coding performance: around -50% rate versus Cool-chic 4.1 for the same visual quality
• Low decoding complexity 1728 MAC / pixel

Attached to this release two bitstreams and the corresponding decoded images, with (wd) and without (mse) --tune=wasserstein.

Below, a subjective tests carried out on the CLIC 2025 test dataset comparing Cool-chic 4.1 (cc_mse), Cool-chic 4.2 with --tune=wasserstein (cc_wd) and H.266 / VVC (vtm). Additional details are available in Perceptually optimised Cool-chic for CLIC 2025, Philippe et al.

🔮 Cool-chic 4.1.0: Revisiting temporal prediction! 🔮

Cool-chic 4.1 focuses on video coding, improving particularly the temporal
prediction through better and lighter sub-pixel motion compensation. This
release is linked to the following paper: Efficient Sub-pixel Motion Compensation in Learned Video Codecs, Ladune et al..

• Replace 2-tap bilinear filtering with sinc-based 8-tap filters
• Decode motion fields at 1/4 of the frame resolution for lighter decoding complexity
• Improved video compression performance: -23.6% rate versus Cool-chic 4.0
• Decrease motion-related complexity by 45%, from 391 to 214 MAC / decoded pixel

Check-out the release history to see previous versions of Cool-chic.

This release does not affect image compression performance which remains identical to v4.0.0

Video compression performance

The Cool-chic page provides comprehensive video rate-distortion results.

Random access results (intra period 32)

Image compression performance

The Cool-chic page provides comprehensive image rate-distortion results.

🎥 Cool-chic 4.0.0: Video is back! 🎥

This release brings back video compression with Cool-chic. Based on the latest Cool-chic video paper. The main improvements are:

• Cool-chic now compresses inter frames with two cool-chic modules: one for the motion and one for the residue
• The motion module training is guided through a RAFT optical flow estimator for more accurate motion.

This release does not affect image compression performance which remains identical to v3.4.1

Image compression performance

Video compression performance (Random Acess, intra period 32)

🎉 Cool-chic 3.4.1: Up to a 3x encoding speed-up! 🎉

This release only affects the encoding speed of Cool-chic, mostly by leveraging torch.compile to run faster.

• Speed-up GPU Encoding up to 3 times! Check-out the detailed encoding speed-up measurements
• Change encoder configuration files with better medium and slow encoding presets

🎉 Cool-chic 3.4.0: 30% less complex! 🎉

This release introduces a new upsampling of the latent variables bringing compression gain. It also allows to rely on smaller Auto-Regressive Modules, lowering the overall decoder complexity significantly.

• 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

Compression performance

Kodak

CLIC20 Pro Valid

JVET Class B

🚀 Cool-chic 3.3: An even faster decoder 🚀

This release improves the speed of Cool-chic CPU-only C decoder and does not affect the compression performance.

Notable improvements:

• Decode a 720p image in 100 ms, 2x faster than Cool-chic 3.2
• Full integerization of the decoder for replicability
• Reduce decoder memory footprint
• Optimized implementation of 3x3 convolutions & fusion of successive 1x1 convolutions