Research map / Video / Video segmentation and tracking
Video Recovery via Learning Variation and Consistency of Images
Zhouyuan Huo, Shangqian Gao, Weidong Cai, Heng Huang
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 l2 smoothness with capped l1 smoothness (capTensor vs l2Tensor) (ablation) | tensor completion · Video recovery D1-D6 | RSE (x10^-2): 2.37 -> 1.30 on D1 | Improved |
| replace l2 with capped l1 in sectional trace norm model (kmsv-l2 -> STN-VCC) (ablation) | STN-VCC · Video recovery D5 Bike | RSE (x10^-2): 1.54 -> 0.96 | Improved |
| replace trace norm with sectional trace norm (capTensor -> STN-VCC) (ablation) | capped l1 video recovery · Video recovery D3 MotherW | RSE (x10^-2): 0.62 -> 0.53 | Improved |
| replace trace norm with sectional trace norm (ablation) | capped l1 video recovery · Video recovery D7 | RSE (x10^-2): 1.42 -> 0.75 | Improved |
| replace trace norm with sectional trace norm (ablation) | capped l1 video recovery · Video recovery D9 | RSE (x10^-2): 0.59 -> 0.57 | No clear effect |
| add sectional trace norm to l2 consistency (l2Tensor -> kmsv-l2) (ablation) | l2 consistency model · Video recovery D3 | RSE (x10^-2): 1.33 -> 0.85 | Improved |
| add sectional trace norm to l2 consistency (l2Tensor -> kmsv-l2) (ablation) | l2 consistency model · Video recovery D12 | RSE (x10^-2): 1.58 -> 1.51 | No clear effect |
| use capped l1 instead of l2 and l1 in toy experiment (ablation) | regularizer · Two consecutive images | difference distribution | Improved |
| minimize squared k-smallest singular values instead of sum of k-smallest (ablation) | low-rank approximation · Image matrix rank 12 | singular values after 12th | Improved |
| vary rank r from 5 to 30 (ablation) | STN-VCC · Sales clip | RSE: best near r = 10 | Mixed |
| vary number of outliers N_eps from 1 to 10000 (ablation) | STN-VCC · Sales clip | RSE | No clear effect |
| use proximal gradient instead of ADMM | STN-VCC · optimization | convergence | Improved |
Closest papers by meaning
- Multi-Matrix Completion: A Novel Framework for Structurally Missing Elements (TPAMI 2025 · similarity 0.87)
- Low Rank Tensor Completion with Poisson Observations (TPAMI 2021 · similarity 0.86)
- Low-Rank Tensor Completion With a New Tensor Nuclear Norm Induced by Invertible Linear Transforms (CVPR 2019 · similarity 0.85)
- High-Rank Matrix Completion and Clustering under Self-Expressive Models (NeurIPS 2016 · similarity 0.85)
- Smooth Robust Tensor Completion for Background/Foreground Separation with Missing Pixels: Novel Algorithm with Convergence Guarantee (JMLR 2022 · similarity 0.85)
- Low-Rank Tensor Completion: A Pseudo-Bayesian Learning Approach (ICCV 2017 · 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.
