Research map / Detection and segmentation / CNN features
OTA: Optimal Transport Assignment for Object Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 per-object assignment with global Optimal Transport Assignment | FCOS with ResNet-50 · COCO val | AP: 38.3 -> 39.2 | Helped |
| add IoU branch (ablation) | OTA-FCOS · COCO val | AP: 39.2 -> 39.6 | Helped |
| add center prior (ablation) | OTA-FCOS with IoU branch · COCO val | AP: 39.6 -> 40.3 | Helped |
| use Dynamic k instead of fixed k (ablation) | OTA-FCOS with IoU branch and center prior · COCO val | AP: 40.3 -> 40.7 | Helped |
| set k to 1, making assignment one-to-one (ablation) | OTA · COCO val | AP: 40.3 -> 36.5 | Hurt |
| vary fixed k from 10 to 20 (ablation) | OTA · COCO val | AP: 40.3 -> 40.1 | No clear effect |
| increase center-prior radius from r=3 to r=5 (ablation) | OTA · COCO val | AP: 40.6 -> 40.7 | Helped |
| increase center-prior radius from r=5 to r=7 (ablation) | OTA · COCO val | AP: 40.7 -> 40.4 | Hurt |
| resolve ambiguous anchors with Min Area before OTA (ablation) | OTA · COCO val | AP: 40.7 -> 40.0 | Hurt |
| resolve ambiguous anchors with Max IoU before OTA (ablation) | OTA · COCO val | AP: 40.7 -> 40.3 | Hurt |
| resolve ambiguous anchors with Min Loss before OTA (ablation) | OTA · COCO val | AP: 40.7 -> 40.3 | Hurt |
| use global OTA rather than PAA assignment | one-stage detector on CrowdHuman · CrowdHuman val | MR: 47.9 -> 46.6 | Helped |
Closest papers by meaning
- Unbalanced Optimal Transport: A Unified Framework for Object Detection (CVPR 2023 · similarity 0.88)
- A Dual Weighting Label Assignment Scheme for Object Detection (CVPR 2022 · similarity 0.87)
- Integrating Diverse Assignment Strategies into DETRs (AAAI 2026 · similarity 0.86)
- LEOD: Label-Efficient Object Detection for Event Cameras (CVPR 2024 · similarity 0.83)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.83)
- DetKDS: Knowledge Distillation Search for Object Detectors (ICML 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.