Research map / Graph learning / Graph algorithms and clustering
To Fold or Not to Fold: Graph Regularized Tensor Train for Visual Data Completion
Le Xu, Lei Cheng, Ngai Wong, Yik‐Chung 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 |
|---|---|---|---|
| not folding the tensor, using graph regularization (GraphTT-VI) instead of folding (ablation) | GraphTT-VI · Image completion, airplane, 90% missing | PSNR (dB): 22.14 -> 23.34 | Improved |
| not folding the tensor, using graph regularization (GraphTT-VI) instead of folding (ablation) | GraphTT-VI · Image completion, facade, 90% missing | PSNR (dB): 20.38 -> 27.14 | Improved |
| not folding the tensor, using graph regularization (GraphTT-VI) instead of folding (ablation) | GraphTT-VI · Image completion, barbara, 90% missing | PSNR (dB): 22.90 -> 24.29 | Improved |
| add graph regularization without folding on TMAC-TT style completion (GraphTT-VI vs TMAC-TT with ket-folding) | TT completion · airplane, 60% missing entries | PSNR (dB): 24.75 -> 29.81 | Improved |
| remove folding from TMAC-TT without adding graph regularization | TMAC-TT · airplane, 60% missing entries | PSNR (dB): 24.75 -> 22.69 | Hurt |
| add TV regularization to folded TT completion (TTC-TV vs TMAC-TT) | TMAC-TT · airplane, 60% missing entries | PSNR (dB): 24.75 -> 23.87 | Hurt |
| use Bayesian VI (GraphTT-VI) instead of optimization with manual parameters (GraphTT-opt) | GraphTT-opt · Image completion, facade, 90% missing | PSNR (dB): 26.10 -> 27.14 | Improved |
| use Bayesian VI (GraphTT-VI) instead of GraphTT-opt | GraphTT-opt · Image completion, barbara, 90% missing | PSNR (dB): 24.38 -> 24.29 | No clear effect |
| increase regularization weight beta0 to 5 from 0.5 (ablation) | GraphTT-opt · synthetic TT data, SNR 10dB, 90% missing | RSE | Hurt |
| increase initial TT rank from 5 to 15 (ablation) | GraphTT-opt · synthetic TT data, SNR 10dB, 90% missing | RSE: 5 -> 15 | Hurt |
| increase initial TT rank from 5 to 15 (ablation) | GraphTT-VI · synthetic TT data, SNR 10dB, 90% missing | RSE: 5 -> 15 | No clear effect |
| update TT core fibers instead of whole TT cores (ablation) | GraphTT-opt · synthetic TT data | RSE and time per iteration | Improved |
Closest papers by meaning
- Low-Rank Tensor Completion with Total Variation for Visual Data Inpainting (AAAI 2017 · similarity 0.89)
- Coarse-To-Fine Tensor Trains for Compact Visual Representations (ICML 2024 · similarity 0.89)
- Simultaneous Visual Data Completion and Denoising Based on Tensor Rank and Total Variation Minimization and Its Primal-Dual Splitting Algorithm (CVPR 2017 · similarity 0.89)
- Fast Tensor Completion via Approximate Richardson Iteration (ICML 2025 · similarity 0.86)
- Knowledge Graph Completion by Intermediate Variables Regularization (NeurIPS 2024 · similarity 0.85)
- Performance Gaps in Multi-view Clustering under the Nested Matrix-Tensor Model (ICLR 2024 · 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.
