Research map / Detection and segmentation / CNN features
One-to-Few Label Assignment for End-to-End Dense Detection
Read the paperCheck 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 o2o assignment with dynamic soft o2f labels for ambiguous anchors | FCOS, ResNet-50 · COCO val, 1× | AP: 36.5 -> 39.0 | Helped |
| Replace o2m assignment with o2f assignment for end-to-end inference | FCOS, ResNet-50 · COCO val, 1× | AP: 17.7 -> 39.0 | Helped |
| Add soft positive supervision to ambiguous anchors instead of adding a second fully positive anchor (ablation) | FCOS · COCO val | AP: 25.9 -> 37.5 | Helped |
| Raise the fixed soft positive degree from t=0.2 to t=0.8 (ablation) | O2F soft-label assignment · COCO val | AP: 37.5 -> 25.6 | Hurt |
| Use seven ambiguous anchors per object rather than thirteen (ablation) | O2F · COCO val, without NMS | AP: 38.4 -> 39.0 | Helped |
| Use a smaller final positive degree for ambiguous anchors (ablation) | O2F · COCO val, without NMS | AP: 39.0 at T min=0.2; 38.0 at T min=0.4 (T max=0.6) | Helped |
| Increase the initial positive degree for ambiguous anchors (ablation) | O2F · COCO val, without NMS | AP: 38.3 at T max=0.4; 39.0 at T max=0.6 (T min=0.2) | Helped |
| Combine classification confidence and regression IoU with a multiplicative matching score rather than an additive score (ablation) | O2F anchor selection · COCO val, without NMS | AP: 37.9 -> 39.0 | Helped |
| Use regression IoU alone for anchor matching (ablation) | O2F anchor selection · COCO val | AP: 10.4 | Hurt |
| Replace dynamic soft-label weights with static Hybrid-Epoch weights (ablation) | FCOS end-to-end detection · COCO val, 1× | AP: 37.3 vs. 39.0 | Hurt |
| Add o2f assignment to CondInst instead of o2o assignment | CondInst, ResNet-50 · COCO val instance segmentation, 1× | AP: 32.8 -> 35.9 | Helped |
Closest papers by meaning
- Dense Distinct Query for End-to-End Object Detection (CVPR 2023 · similarity 0.88)
- End-to-End Object Detection With Fully Convolutional Network (CVPR 2021 · similarity 0.88)
- What Makes for End-to-End Object Detection? (ICML 2021 · similarity 0.87)
- Dynamic Coarse-To-Fine Learning for Oriented Tiny Object Detection (CVPR 2023 · similarity 0.86)
- Integrating Diverse Assignment Strategies into DETRs (AAAI 2026 · similarity 0.86)
- Detect, Replace, Refine: Deep Structured Prediction for Pixel Wise Labeling (CVPR 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.