Research map / Detection and segmentation / CNN features
Anchor-Intermediate Detector: Decoupling and Coupling Bounding Boxes for Accurate Object Detection
Yilong Lv, Min Li, Yujie He, Shaopeng Li, Zhuzhen He, Aitao Yang
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 |
|---|---|---|---|
| attach corner-aware head in parallel after FPN rather than inside cls or reg branch (ablation) | AID on RetinaNet R101 · COCO val2017 subset | AP: 39.4 -> 39.8 | Improved |
| increase corner loss weight lambda from 0.3 to 1 (ablation) | AID on RetinaNet R101 · COCO val2017 subset | AP: 40.1 -> 38.2 | Hurt |
| add CoCl score (classification times corner score) in post-processing (ablation) | AID · COCO val2017 subset | AP: 38.5 -> 39.8 | Improved |
| add BDC strategy on top of CoCl score (ablation) | AID · COCO val2017 subset | AP: 39.8 -> 40.0 | Improved |
| average all corner points in BDC couple step (ablation) | BDC · COCO val2017 subset | AP: 38.5 -> 36.0 | Hurt |
| use score-threshold corner selection instead of max or top-n (ablation) | BDC · COCO val2017 subset | AP: 39.7 -> 40.0 | Improved |
| use avg, max or min of two corner heatmaps in CoCl score (ablation) | CoCl score · COCO val2017 subset | AP: 39.8 / 39.6 / 39.6 | No clear effect |
| use only classification score without corner score in CoCl exponent form (ablation) | CoCl score · COCO val2017 subset | AP: 39.9 -> 28.5 | Hurt |
| replace FPN with PAFPN (ablation) | AID · COCO val2017 subset | AP: 40.0 -> 40.1 | Improved |
| replace MaxIoUAssigner with ATSSAssigner (ablation) | AID · COCO val2017 subset | AP: 40.0 -> 40.6 | Improved |
| add AID to RetinaNet with ResNeXt-101-64x4d | RetinaNet · COCO test-dev 1x | AP: 41.4 -> 41.7 | Improved |
| add AID to GFL | GFL · COCO test-dev | AP: +0.5 to +1.2 | Improved |
Closest papers by meaning
- AABO: Adaptive Anchor Box Optimization for Object Detection via Bayesian Sub-sampling (ECCV 2020 · similarity 0.88)
- Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (CVPR 2020 · similarity 0.88)
- Soft Anchor-Point Object Detection (ECCV 2020 · similarity 0.88)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · similarity 0.86)
- Theoretically Achieving Continuous Representation of Oriented Bounding Boxes (CVPR 2024 · similarity 0.85)
- Enhancing 3D Object Detection with 2D Detection-Guided Query Anchors (CVPR 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.
