TensorFlow Implementation of Checkerboard Context Model (He et al. CVPR 2021)

[Nikolai10]

Dear TFC-Community,

we have recently been working on a TF implementation of Checkerboard Context Model for Efficient Learned Image Compression (He et al. CVPR 2021): https://github.com/Nikolai10/Checkerboard-Context-Model.

The Checkerboard Context Model is an important pillar for efficient learned image compression. It is also a key ingredient to ELIC, a close to the state-of-the-art image compression method.

This project is joint work with Yang Zhang (@Yango4you), who conducted his Forschungspraxis at TUM with me.

Based on our observations, MS2020 + Checkerboard allows 1.88x decoding speedup over MS2020 + ChARM, while only sacrificing 0.26dB in performance. This is achieved despite using a significantly smaller model size (26.4M vs. 116.3M).

In our implementation we have taken some care to closely follow the naming conventions presented by Minnen et al., without changing the overall structure. We have also included a Google Colab demo, which you can find here.

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

Kind regards,
Nikolai