Transform Network Architectures for Deep Learning Based End-to-End Image/Video Coding in Subsampled Color Spaces

Most of the existing deep learning based end-to-end image/video coding (DLEC) architectures are designed for non-subsampled RGB color format. However, in order to achieve a superior performance, many state-of-the-art block-based compression standards such as High Efficiency Video Coding (HEVC/H.265) and Versatile (VVC/H.266) primarily YUV 4:2:0 format, where U V components subsampled by considering human visual system. This paper investigates various DLEC designs support format comparing their performance against main profiles HEVC VVC under common evaluation framework. Moreover, new transform network architecture is proposed improve efficiency data. The experimental results on datasets show that significantly outperforms naive extensions achieves about 10% average BD-rate improvement over intra-frame HEVC.