Scalable Image Coding for Humans and Machines Using Feature Differences (特徴量差分を用いた視聴と画像認識のための階層画像符号化)

With the advancement of deep learning, the demand for scalable image coding that compresses images for both human vision and image recognition tasks has been increasing. Previous research proposed a method that connects two LIC models using a feature fusion network. We improve this method by incorporating the differences of the features.

In Picture Coding Symposium Japan 2024