Learned Multi-Resolution Variable-Rate Image Compression With Octave-Based Residual Blocks

Recently deep learning-based image compression has shown the potential to outperform traditional codecs. However, most existing methods train multiple networks for bit rates, which increase implementation complexity. In this paper, we propose a new variable-rate framework, employs generalized octave convolutions (GoConv) and transposed-convolutions (GoTConv) with built-in divisive normalization (GDN) inverse GDN (IGDN) layers. Novel GoConv- GoTConv-based residual blocks are also developed in encoder decoder networks. Our scheme uses stochastic rounding-based scalar quantization. To further improve performance, encode between input reconstructed from network as an enhancement layer. enable single model operate different rates learn multi-rate features, objective function is introduced. Experimental results show that proposed framework trained outperforms standard codecs such H.265/HEVC-based BPG state-of-the-art methods.