Rate-Distortion Optimized Encoding for Deep Image Compression
چکیده
Deep-learned variational auto-encoders (VAE) have shown remarkable capabilities for lossy image compression. These neural networks typically employ non-linear convolutional layers finding a compressible representation of the input image. Advanced techniques such as vector quantization, context-adaptive arithmetic coding and variable-rate compression been implemented in these auto-encoders. Notably, rely on an end-to-end approach, which fundamentally differs from hybrid, block-based video systems. Therefore, signal-dependent encoder optimizations not thoroughly investigated VAEs yet. However, rate-distortion optimized encoding heavily determines performance state-of-the-art codecs. Designing non-linear, multi-layered requires to understand relationship between bit allocation features distortion. this paper examines VAE. In particular, one demonstrates that trained network finds with near-optimal across channels. Furthermore, approximates distortion quantization by higher-order polynomial, whose coefficients can be robustly estimated. Based considerations, authors investigate algorithm Lagrange optimization, significantly improves efficiency.