Research map / Video / Video segmentation and tracking
Motion Information Propagation for Neural Video Compression
Linfeng Qi, Jiahao Li, Bin Li, Houqiang Li, Yan Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace DCVC-HEM with DCVC-MIP | neural video compression · Average across UVG, MCL-JCV, HEVC B/C/D/E and HEVC RGB | BD-Rate (%) PSNR: -4.7 -> -16.6 | Improved |
| replace DCVC-HEM with DCVC-MIP | neural video compression · Average across UVG, MCL-JCV, HEVC B/C/D/E and HEVC RGB | BD-Rate (%) MS-SSIM: -46.4 -> -50.8 | Improved |
| add offset diversity (ablation) | DCVC-MIP context generation · average benchmark performance | BD-Rate (%): 0.0 -> -2.6 | Improved |
| add motion guidance (ablation) | DCVC-MIP context generation · average benchmark performance | BD-Rate (%): -2.6 -> -5.9 | Improved |
| add motion condition (ablation) | DCVC-MIP motion coding · average benchmark performance | BD-Rate (%): -5.9 -> -9.4 | Improved |
| add transformer-based context refinement (ablation) | DCVC-MIP context generation · average benchmark performance | BD-Rate (%): -9.4 -> -12.9 | Improved |
| use motion feature Gt rather than decoded motion flow as guidance (ablation) | DCVC-MIP · average benchmark performance | BD-Rate (%): -2.4 -> -3.2 | Improved |
| use propagated feature Ht−1 as motion condition (ablation) | DCVC-MIP motion coding · average benchmark performance | BD-Rate (%): -3.2 -> -6.9 | Improved |
| use residual offsets Rt−1 rather than feature Ht−1 as motion condition (ablation) | DCVC-MIP motion coding · average benchmark performance | BD-Rate (%): -6.9 -> -4.6 | Hurt |
| apply transformer-based context refinement at 1/4 scale (ablation) | DCVC-MIP context generation · average benchmark performance | BD-Rate (%): 0.0 -> -3.2 | Improved |
| apply RSTB to the entropy model instead of TCR at 1/4 scale (ablation) | DCVC-MIP entropy model · average benchmark performance | BD-Rate (%): -3.2 -> -1.4 | Hurt |
Closest papers by meaning
- Neural Video Compression with Context Modulation (CVPR 2025 · similarity 0.89)
- High Resolution Neural Video Coding with Bi-directional Confidence-Guided Reference Information Modeling (CVPR 2026 · similarity 0.89)
- Neural B-frame Video Compression with Bi-directional Reference Harmonization (NeurIPS 2025 · similarity 0.87)
- Augmented Deep Contexts for Spatially Embedded Video Coding (CVPR 2025 · similarity 0.86)
- Towards Practical Real-Time Neural Video Compression (CVPR 2025 · similarity 0.85)
- Neural Video Compression with Feature Modulation (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
