PyTorch implementation of PerCo (Towards Image Compression with Perfect Realism at Ultra-Low Bitrates, ICLR 2024)

[Nikolai10]

Dear TFC-Community,

I have recently been working on a PyTorch implementation of PerCo (Careil et al. ICLR 2024):
https://github.com/Nikolai10/PerCo.

PerCo considers ultra-low bitrate image compression with powerful pre-trained text-to-image models (e.g. GLIDE, Stable Diffusion), thereby achieving unmatched visual quality, while featuring a simple and elegant design.

It shares some connections to:

• Generative Adversarial Networks for Extreme Learned Image Compression (Agustsson et al. ICCV 2019) and
• Deep Generative Models for Distribution-Preserving Lossy Compression (Tschannen et al. NIPS 2018).

While the code base is written in PyTorch, the compress & decompress functionality (CLI interface) is adopted from TFC - so TFC users should be able to experiment with it as well.

I hope this work will be useful for further exploration and learning.

Kind regards,
Nikolai