Research map / Video / Video segmentation and tracking
NVC-1B: Scaling up Neural Video Coding Models
CP Tang, Xihua Sheng, Li Li, Dong Liu, Feng Wu
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 |
|---|---|---|---|
| scale motion encoder-decoder from 52M to 96M (ablation) | DCVC-SDD · HEVC Class C | BD-rate gain (%): 15.4% -> 1.5% | Hurt |
| scale motion entropy model (M1 to M4) (ablation) | DCVC-SDD · HEVC Class C | BD-rate gain (%): 15.4% -> 16.0% | Improved |
| scale contextual encoder-decoder from 50M to 90M (ablation) | DCVC-SDD · HEVC Class C | BD-rate gain (%): 10.6% -> 14.8% | Improved |
| scale contextual entropy model from 90M to 212M (ablation) | DCVC-SDD · HEVC Class C | BD-rate gain (%): additional 0.8% | Improved |
| scale temporal context mining from 26M to 75M (ablation) | DCVC-SDD · HEVC Class D | BD-rate gain (%): 5.1% -> 12.0% | Improved |
| scale temporal context mining on top of M12 (M18) (ablation) | DCVC-SDD · HEVC Class D | BD-rate gain (%): 14.2% -> 21.2% | Improved |
| use mixed CNN-Transformer instead of CNN at similar size (ablation) | DCVC-SDD scaled model · HEVC Class C | BD-rate gain (%): 15.6% -> 9.4% | Hurt |
| insert Swin layers into temporal context mining (ablation) | M20 · HEVC Class C | BD-rate gain (%): 9.4% -> 2.9% | Hurt |
| replace convolutions with Transformer-only entropy model and temporal context (ablation) | DCVC-SDD scaled model · RGB-PSNR | BD-rate gain (%): PSNR loss of more than 4dB | Hurt |
| scale up to NVC-1B (1.36B parameters) | DCVC-SDD · RGB average BD-rate vs VTM | BD-rate gain (%): -12.7 -> -25.1 | Improved |
| scale up to NVC-1B | DCVC-SDD · YUV average BD-rate vs VTM | BD-rate gain (%): -8.4 -> -20.5 | Improved |
| scale up to NVC-1B | DCVC-SDD · 1080P encoding time | BD-rate gain (%): 0.89 s -> 4.44 s | Hurt |
Closest papers by meaning
- Towards Practical Real-Time Neural Video Compression (CVPR 2025 · similarity 0.89)
- Neural Video Compression with Feature Modulation (CVPR 2024 · similarity 0.89)
- NVRC: Neural Video Representation Compression (NeurIPS 2024 · similarity 0.88)
- Neural Video Compression with Context Modulation (CVPR 2025 · similarity 0.87)
- PNVC: Towards Practical INR-based Video Compression (AAAI 2025 · similarity 0.87)
- Combining Frame and GOP Embeddings for Neural Video Representation (CVPR 2024 · similarity 0.87)
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.
