Research map / Video / Video segmentation and tracking
CANF-VC: Conditional Augmented Normalizing Flows for Video Compression
Yung-Han Ho, Chih-Peng Chang, Peng-Yu Chen, Alessandro Gnutti, Wen-Hsiao Peng
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 |
|---|---|---|---|
| use conditional inter-frame coding instead of residual coding, no temporal prior (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -21.8% -> -28.9% | Improved |
| use conditional inter-frame coding instead of residual coding, with temporal prior (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -24.3% -> -33.1% | Improved |
| add temporal prior to residual coding (ablation) | ANFIC residual coding · UVG | BD-rate (PSNR-RGB, lower is better): -21.8% -> -24.3% | Improved |
| add temporal prior to conditional inter-frame coding (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -28.9% -> -33.1% | Improved |
| use conditional motion coding with extrapolated flow instead of predictive motion coding (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -35.3% -> -42.5% | Improved |
| use conditional motion coding with extrapolated flow instead of intra motion coding (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -33.4% -> -42.5% | Improved |
| condition motion coding on previous decoded flow instead of extrapolated flow (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -42.5% -> -35.2% | Hurt |
| use 2-step instead of 1-step CANF inter-frame coder (ablation) | CANF-VC with DVC motion coder · UVG | BD-rate (PSNR-RGB, lower is better): -31.4% -> -33.1% | Improved |
| use 2-step instead of 1-step CANF motion coder (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -38.1% -> -42.5% | Improved |
| use 3-step instead of 2-step CANF in both coders (ablation) | CANF-VC · UVG | BD-rate (PSNR-RGB, lower is better): -42.5% -> -31.4% | Hurt |
| add conditional motion coding (CANF-VC vs CANF-VC-) (ablation) | CANF-VC · UVG (GOP 32) | BD-rate (PSNR-RGB, lower is better): -31.1 -> -35.9 | Improved |
Later papers that compared with this method
2 later accepted papers report CANF-VC as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- VNVC: A Versatile Neural Video Coding Framework for Efficient Human-Machine Vision (TPAMI 2024 · better on 3 of 3 shared results)
- Scene Matters: Model-based Deep Video Compression (ICCV 2023 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Hierarchical B-Frame Video Coding Using Two-Layer CANF Without Motion Coding (CVPR 2023 · similarity 0.90)
- VC-VAE: Leveraging Video Codecs for Training-Efficient and High-Fidelity Video VAE (ECCV 2026 · similarity 0.86)
- Insights From Generative Modeling for Neural Video Compression (TPAMI 2023 · similarity 0.86)
- Neural Video Compression with Feature Modulation (CVPR 2024 · similarity 0.84)
- Generative Latent Coding for Ultra-Low Bitrate Image Compression (CVPR 2024 · similarity 0.84)
- Implicit Motion Function (CVPR 2024 · similarity 0.84)
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.
