# Cool-chic: Leveraging overfitting to design a low complexity neural image codec _by Théo Ladune (Orange) - 2023.10.18_ ###### tags: `VAADER` `Seminar` ![](https://hackmd.io/_uploads/rJqwIlJkp.bmp) ## Abstract Most neural image and video codecs rely on autoencoders, performing up to a million multiplications per decoded pixel. Despite state-of-the-art compression performance, this important complexity hinders the adoption of neural codecs. This talk presents Cool-chic, a neural image codec leveraging overfitting to obtain compelling compression performance while requiring only 1 000 multiplications per decoded pixel. Cool-chic implementation is available at https://orange-opensource.github.io/Cool-Chic/.