Research map / Detection and segmentation / Open-set and few-shot detection
MessyTable: Instance Association in Multiple Camera Views
Zhongang Cai, Junzhe Zhang, Daxuan Ren, Cunjun Yu, Haiyu Zhao, Shuai Yi, Chai Kiat Yeo, Chen Change Loy
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 a separate surrounding branch (ASNet) instead of an appearance-only triplet network (ablation) | TripletNet · MessyTable (all scenes) | AP: 0.467 -> 0.524 | Improved |
| naively zoom out the input crop to include surroundings (ablation) | TripletNet · MessyTable (all scenes) | AP: 0.467 -> 0.430 | Hurt |
| add the epipolar soft constraint (ablation) | ASNet · MessyTable (all scenes) | AP: 0.524 -> 0.577 | Improved |
| add the epipolar soft constraint (ablation) | ASNet · MessyTable (all scenes) | IPAA-100: 0.170 -> 0.219 | Improved |
| use a deeper backbone (ablation) | MatchNet · MessyTable (all scenes) | AP: 0.193 -> 0.138 | Hurt |
| use a deeper backbone (ablation) | DeepCompare · MessyTable (all scenes) | AP: 0.202 -> 0.129 | Hurt |
| use a deeper backbone (ablation) | DeepDesc · MessyTable (all scenes) | AP: 0.090 -> 0.171 | Improved |
| use triplet architecture instead of Siamese patch-matching models | MatchNet / DeepCompare / DeepDesc · MessyTable (all scenes) | AP: 0.467 vs 0.193, 0.202, 0.090 | Improved |
| test on harder scenes (Medium and Hard vs Easy) | ASNet+ESC · MessyTable (Easy/Medium/Hard) | AP: 0.709 -> 0.594 -> 0.457 | Hurt |
| pretrain on MessyTable instead of ImageNet | ASNet · MPII MK | AP: 0.881 -> 0.905 | Improved |
| pretrain on MessyTable instead of ImageNet | ASNet · EPFL MVMC | AP: 0.950 -> 0.969 | Improved |
| pretrain on MessyTable instead of ImageNet | ASNet · WILDTRACK | AP: 0.718 -> 0.734 | Improved |
Closest papers by meaning
- Learning to Generate Multiple Objects from Dense and Occluded Layouts (NeurIPS 2026 · similarity 0.85)
- ORIDa: Object-centric Real-world Image Composition Dataset (CVPR 2025 · similarity 0.84)
- Common Inpainted Objects In-N-Out of Context (CVPR 2026 · similarity 0.83)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.83)
- v-CLR: View-Consistent Learning for Open-World Instance Segmentation (CVPR 2025 · similarity 0.83)
- TABLET: A Large-Scale Dataset for Robust Visual Table Understanding (ICLR 2026 · similarity 0.83)
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.
