Research map / Video / Video segmentation and tracking
Efficient Meta-Tuning for Content-Aware Neural Video Delivery
Xiaoqi Li, Jiaming Liu, Shizun Wang, Cheng Lyu, Ming Lu, Yurong Chen, Anbang Yao, Yandong Guo, Shanghang Zhang
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 meta-tuning (meta-learned init, S setting) instead of per-chunk training from scratch | ESPCN content-aware SR · inter-45s x2 | PSNR / time: 38.95 -> 39.08; 11.2h -> 1.2m | Improved |
| use meta-learned init instead of pretrained init (ablation) | EMT · sport-45s x2 (S) | training time: 4.2m -> 1.2m | Improved |
| use meta-learned init instead of pretrained init (ablation) | EMT · inter-45s x2 (S) | training time: 18.4m -> 1.2m | Improved |
| add challenging patch sampling (ablation) | EMT · inter-45s x2 (S) | training time: 9.7m -> 1.2m | Improved |
| add challenging patch sampling (ablation) | EMT · sport-45s x2 (S) | training time: 2.7m -> 1.2m | Improved |
| update all parameters for first chunk (p1 20 -> 100) (ablation) | EMT Ours(S) · inter-45s x2 | PSNR: 39.08 -> 39.19 | Improved |
| raise p2 from 1 to 10 (ablation) | EMT Ours(S) · sport-45s x2 | PSNR: 46.11 -> 46.23 | Improved |
| vary number of meta-learning tasks (10, 15, 20) (ablation) | EMT Ours(S) · inter-45s x2 | PSNR: 39.06 / 39.08 / 39.07 | No clear effect |
| increase meta-learning frames from 10 to 150 (ablation) | EMT Ours(S) · sport-45s x2 | PSNR: 45.94 -> 46.20 | Mixed |
| apply EMT to x4 scaling instead of C1-n | ESPCN · sport-45s x4 | PSNR: 37.21 -> 37.22 | No clear effect |
| apply EMT to x3 scaling | ESPCN · inter-45s x3 | PSNR: 32.19 -> 32.55 | Improved |
| apply EMT with SRCNN backbone | SRCNN · game-45s | PSNR: 35.37 -> 35.64 | Improved |
Closest papers by meaning
- MetaNeRV: Meta Neural Representations for Videos with Spatial-Temporal Guidance (AAAI 2025 · similarity 0.84)
- HyperNVD: Accelerating Neural Video Decomposition via Hypernetworks (CVPR 2025 · similarity 0.83)
- DeNC: Unleash Neural Codecs in Video Streaming with Diffusion Enhancement (AAAI 2025 · similarity 0.82)
- SNED: Superposition Network Architecture Search for Efficient Video Diffusion Model (CVPR 2024 · similarity 0.82)
- Unleashing the Power of Meta-tuning for Few-shot Generalization Through Sparse Interpolated Experts (ICML 2024 · similarity 0.82)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.82)
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.
