Research map / Video / Video segmentation and tracking
Scene Matters: Model-based Deep Video Compression
Lv Tang, Xinfeng Zhang, Gai Zhang, Xiaoqi Ma
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 |
|---|---|---|---|
| add context-related spatial positional embedding (CRSPE) to baseline video INR (ablation) | video INR baseline · Surveillance-1 | BDBR vs H.266 (%): 13.82 -> 3.15 | Improved |
| add frequency domain supervision (FDS) (ablation) | MVC with CRSPE · Surveillance-1 | BDBR vs H.266 (%): 3.15 -> -3.53 | Improved |
| add scene flow constrain mechanism (SFCM) (ablation) | MVC with CRSPE+FDS · Surveillance-1 | BDBR vs H.266 (%): -3.53 -> -18.65 | Improved |
| add temporal contrastive loss (TCL) (ablation) | MVC with CRSPE+FDS+SFCM · Surveillance-1 | BDBR vs H.266 (%): -18.65 -> -22.97 | Improved |
| add flow regularization to SFCM (ablation) | MVC · Conference | BDBR vs H.266 (%): 4.33 -> -4.21 | Improved |
| add entropy loss L_ent on regularization map (ablation) | MVC SFCM · Surveillance-1 | BDBR vs H.266 (%): -16.69 -> -18.65 | Improved |
| add prior Gaussian distribution weighting to TCL (ablation) | MVC TCL · Surveillance-1 | BDBR vs H.266 (%): -18.12 -> -22.97 | Improved |
| reduce TCL frame window L from 80 to 40 (ablation) | MVC TCL · Surveillance-1 | BDBR vs H.266 (%): -22.97 -> -19.85 | Hurt |
| increase TCL frame window L from 80 to 120 (ablation) | MVC TCL · Surveillance-1 | BDBR vs H.266 (%): -22.97 -> -21.97 | No clear effect |
| transmit CRSPE embedding | MVC 720p videos · 720p videos | bitrate and PSNR: extra 10% Bpp for about 0.9dB PSNR | Mixed |
| use multi-GPU parallel decoding | MVC · 600 1080p frames decoding | decoding speed: about 132 s on one GPU; real-time 35 FPS with more GPUs | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CANFVC — CANF-VC: Conditional Augmented Normalizing Flows for Video Compression (ECCV 2022) · 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
- Deep Contextual Video Compression (NeurIPS 2021 · similarity 0.89)
- Augmented Deep Contexts for Spatially Embedded Video Coding (CVPR 2025 · similarity 0.89)
- Video Compression With Rate-Distortion Autoencoders (ICCV 2019 · similarity 0.88)
- Video Compression With Entropy-Constrained Neural Representations (CVPR 2023 · similarity 0.88)
- Neural Video Compression with Context Modulation (CVPR 2025 · similarity 0.87)
- FLAVC: Learned Video Compression with Feature Level Attention (CVPR 2025 · 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.
