MRT: Learning Compact Representations with Mixed RWKV-Transformer for Extreme Image Compression

Recent advancements in extreme image compression have demonstrated that converting pixel data into highly compact latent representations can enhance coding efficiency. Traditional methods often rely on convolutional neural networks (CNNs) or Swin Transformers, which maintain significant spatial redundancy, limiting compression performance. The proposed Mixed RWKV-Transformer (MRT) architecture encodes images into compact 1-D latent representations by integrating the strengths of RWKV and Transformer models, capturing global dependencies and local redundancies effectively.