DCVC-MB: Neural B-Frame Video Compression using State Space Models

Jul 17, 2026

Researchers have introduced DCVC-Mamba (DCVC-MB), a neural video codec framework designed for B-frame coding. The approach leverages an IBP frame strategy, a spatio-temporal fusion model based on state-space models for bidirectional temporal prediction, and an entropy-aware skipping mechanism to improve compression efficiency. Experimental results show that DCVC-MB achieves BD-rate reductions of up to 8.98% compared to previous neural video codecs, and outperforms traditional codecs VTM-19.0-LDP and VTM-19.0-RA by up to 30.45% and 1.81%, respectively.

Why it matters: This work demonstrates a notable advance in neural video compression by introducing state-space models for bidirectional prediction, resulting in significant compression improvements over both neural and traditional codecs.

Full story at: arXiv Computer Vision ↗