Learned Image Compression

Learned image compression (LIC) aims to surpass traditional methods by using deep learning models to achieve superior rate-distortion performance in image encoding and decoding. Current research focuses on improving LIC efficiency through architectural innovations, such as incorporating transformers and convolutional neural networks, and refining entropy models to better capture spatial and channel-wise dependencies within image data. These advancements hold significant promise for reducing storage needs and improving transmission speeds for various image-based applications, impacting fields ranging from medical imaging to multimedia streaming.