A survey of learning-based end-to-end video compression

With the increase of multimedia data and the emergence of intelligent application scenarios such as virtual reality, video compression faces demand for higher resolution and more diverse video data. Compression methods based on end-to end learning have shown great flexibility and certain superiority. Although there are still challenges in computational complexity, artificial intelligence technology has injected more vitality into video compression. The continuous development of visual-language models, artificial intelligence-generated content, and generative models may provide a revolutionary development for compression. Considering these factors, we review new research work and influential articles. Specifically, this paper introduces the development of the video coding group and briefly outlines learning-based image compression methods (intra-frame coding). In particular, we review video compression on different coding frameworks, such as residual and context (inter-frame coding). Finally, we discuss possible future research directions on video compression and the challenges they may face.