Image Compression Model

Learned image compression (LIC) aims to improve upon traditional methods by leveraging deep learning to achieve better rate-distortion performance and perceptual quality. Current research focuses on enhancing existing models, such as variational autoencoders (VAEs) and convolutional neural networks (CNNs), often incorporating transformers or wavelet transforms to better capture image features and handle high-frequency information. These advancements are driven by the need for efficient image transmission and storage in various applications, including machine vision and large language model inference, where improved compression can significantly reduce computational costs and improve speed.