本次简单讲一下Cheng 2020，也是经典的baseline之一，作者也开源了，推荐阅读。 InfoTitle: Learned Image Compression with Discretized Gaussian Mixture Likelihoods and Attention ModulesAuthor: Zhengxue ChengPub: CVPR 2020Repo: http

InfoTitle: Learned Image Compression with Dictionary-based Entropy ModelAuthor: Jingbo LuPub: CVPR 2025Repo: https://github.com/LabShuHangGU/DCAEKeyword: Dictionary-based, Cross Attention Motivation现

InfoTitle: Variational Image Compression with a Scale HyperpriorAuthor: Johannes BallePub: ICLR 2018Keyword: Hyperprior Motivation 已有研究（例如Ballel2017的工作）还是用的变换+量化的形式来压缩。然而这些方法在量化时往往忽略了图像中的空间依赖性，导致压缩性能

InfoTitle: End-to-end optimized image compressionAuthor: Johannes BallePub: ICLR 2017Repo: https://github.com/tensorflow/compression Motivation 联合优化率（rate）和失真（distortion）是困难的，现有算法例如JPEG通常是线性变换、均匀量化和熵