Research map / Detection and segmentation / CNN features
End-to-End Object Detection With Fully Convolutional Network
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 one-to-many assignment with POTO one-to-one assignment | FCOS · COCO val | mAP without NMS: 12.1 -> 38.0 | Helped |
| Add 3DMF to POTO (ablation) | POTO detector · COCO val | mAP without NMS: 38.0 -> 39.8 | Helped |
| Add auxiliary one-to-many loss to POTO+3DMF (ablation) | POTO+3DMF · COCO val | mAP without NMS: 39.8 -> 41.1 | Helped |
| Use prediction-aware multiplicative classification-regression fusion instead of additive fusion (ablation) | POTO matching · COCO val | mAP without NMS: 29.3 -> 38.0 | Helped |
| Combine classification and regression quality rather than using classification alone (ablation) | POTO matching · COCO val | mAP without NMS: 33.5 -> 38.0 | Helped |
| Use classification and regression quality together rather than regression alone (ablation) | POTO matching · COCO val | mAP without NMS: 11.8 -> 38.0 | Helped |
| Use center sampling instead of global assignment candidates (ablation) | POTO matching · COCO val | mAP without NMS: 37.3 -> 38.0 | Helped |
| Use center sampling instead of inside-box assignment candidates (ablation) | POTO matching · COCO val | mAP without NMS: 37.4 -> 38.0 | Helped |
| Add two deformable convolutions in the detection head (ablation) | Ours · COCO val | mAP: 41.5 -> 43.5 | Helped |
| Use 3DMF as an end-to-end trainable module rather than post-processing (ablation) | FCOS detector with 3DMF · COCO val | mAP: +1.1% mAP | Helped |
| Use 3DMF across adjacent scales and local spatial neighborhoods (ablation) | POTO+3DMF detector · COCO val | mAP without NMS: 39.2 -> 39.8 | Helped |
| Add auxiliary prediction-aware quality assignment instead of no auxiliary loss (ablation) | POTO+3DMF · COCO val | mAP without NMS: 39.8 -> 41.1 | Helped |
Closest papers by meaning
- What Makes for End-to-End Object Detection? (ICML 2021 · similarity 0.89)
- Dense Distinct Query for End-to-End Object Detection (CVPR 2023 · similarity 0.89)
- One-to-Few Label Assignment for End-to-End Dense Detection (CVPR 2023 · similarity 0.88)
- Learning Non-Maximum Suppression (CVPR 2017 · similarity 0.88)
- PSRR-MaxpoolNMS: Pyramid Shifted MaxpoolNMS With Relationship Recovery (CVPR 2021 · similarity 0.86)
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.84)
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.