Design of Deep Video Compression Model Enabling Variable Bit Rate Coding

Recently, deep learning has inspired a shift from traditional coding to Deep Video Compression (DVC). Early DVC systems used conventional residual coding, but newer methods with superior conditional entropy coding have surpassed them. Despite advancements, DVC requires multi-bitrate encoding, traditionally managed by training multiple models or different QPs. Our proposed framework achieves multi-rate encoding with a single model, significantly reducing BD-Rate by 32.406% and 66.302% for PSNR and MS-SSIM, respectively, compared to the x265 system.