Research map / Detection and segmentation / CNN features
Probabilistic Anchor Assignment with IoU Prediction for Object Detection
Kang Kim, Hee Seok Lee
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 |
|---|---|---|---|
| replace IoU-based hard assignment with probabilistic anchor assignment (ablation) | IoU-based assignment with IoU pred. · COCO minival (ResNet-50) | AP: 36.0 -> 40.9 | Improved |
| add IoU prediction head and unified score in NMS (ablation) | PAA · COCO minival (ResNet-50) | AP: 39.9 -> 40.9 | Improved |
| add score voting (ablation) | PAA with IoU prediction · COCO minival (ResNet-50) | AP: 40.9 -> 41.1 | Improved |
| use centerness prediction instead of IoU prediction (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.9 -> 39.8 | Hurt |
| use IoU prediction instead of centerness prediction (ablation) | ATSS · COCO minival (ResNet-50) | AP: 39.4 -> 39.8 | Improved |
| add IoU prediction to IoU-based assignment (ablation) | IoU-based assignment · COCO minival (ResNet-50) | AP: 34.6 -> 36.0 | Improved |
| replace probabilistic separation with fixed number of positives (FNP 5) (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.9 -> 39.5 | Hurt |
| replace probabilistic separation with fixed number of positives (FNP 10) (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.9 -> 40.1 | Hurt |
| replace probabilistic separation with fixed score threshold (FSR >0.1) (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.9 -> 23.8 | Hurt |
| fixed score threshold FSR >0.3 (ablation) | PAA · COCO minival (ResNet-50) | AP: training failed | Hurt |
| vary separation boundary scheme (a) to (c) (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.5 -> 40.9 | No clear effect |
| vary topk K from 5 to 9 to 25 (ablation) | PAA · COCO minival (ResNet-50) | AP: 40.5 -> 40.9 | No clear effect |
Closest papers by meaning
- Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (CVPR 2020 · similarity 0.91)
- Learning to Match Anchors for Visual Object Detection (TPAMI 2021 · similarity 0.90)
- Learning From Noisy Anchors for One-Stage Object Detection (CVPR 2020 · similarity 0.89)
- FreeAnchor: Learning to Match Anchors for Visual Object Detection (NeurIPS 2019 · similarity 0.89)
- Multiple Anchor Learning for Visual Object Detection (CVPR 2020 · similarity 0.89)
- Region Proposal by Guided Anchoring (CVPR 2019 · similarity 0.89)
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.
