ELIC, short for Efficient Learned Image Compression, is a deep learning-based framework designed for lossy image compression that achieves high compression efficiency alongside fast computational speed. The architecture enhances rate-distortion performance by employing a spatial-channel contextual adaptive model, which groups latent representation channels unevenly to exploit the natural energy compaction of feature representations. By combining efficient transform network designs with parallelizable probability estimation, ELIC improves encoding and decoding throughput while also enabling functionalities such as rapid preview and progressive decoding.
