Research map / Detection and segmentation / CNN features
Learning to Match Anchors for Visual Object Detection
Xiaosong Zhang, Fang Wan, Chang Liu, Xiangyang Ji, Qixiang Ye
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 assignment with FreeAnchor MLE matching | RetinaNet ResNet-50 · COCO | AP: 35.7 -> 38.7 | Improved |
| replace IoU-based assignment with FreeAnchor MLE matching | RetinaNet ResNet-101 · COCO | AP: 37.8 -> 40.9 | Improved |
| add FreeAnchor learning-to-match | RetinaNet ResNet-50 · COCO val | NMS recall (NR): 81.8 -> 83.8 | Improved |
| add FreeAnchor learning-to-match | RetinaNet ResNet-50 · COCO val | NR90: 51.3 -> 53.1 | Improved |
| add FreeAnchor learning-to-match | RetinaNet · COCO | training time (h): 5.02h -> 5.27h | No clear effect |
| apply FreeAnchor to slender objects | RetinaNet · COCO slender objects | AP | Improved |
| apply FreeAnchor to square objects | RetinaNet · COCO square objects | AP | No clear effect |
| increase number of objects per image (crowdedness) | FreeAnchor vs RetinaNet · COCO | AP gap | Improved |
| set anchor bag size to 50 among {40, 50, 60, 100} (ablation) | FreeAnchor · COCO | AP: bag size 50 best | Improved |
| set background IoU threshold to 0.6 among {0.5, 0.6, 0.7} (ablation) | FreeAnchor · COCO | AP: 0.6 best | Improved |
| extend scale range from [640, 800] to [480, 960] (ablation) | FreeAnchor ResNeXt-101 · COCO test-dev | AP: 44.9 -> 46.0 | Improved |
| add multi-scale testing over six scales (ablation) | FreeAnchor* ResNeXt-101 · COCO test-dev | AP: 46.0 -> 47.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FCOS — FCOS: Fully Convolutional One-Stage Object Detection (ICCV 2019) · better on 3 of 3 shared results
- CornerNet — CornerNet: Detecting Objects as Paired Keypoints (ECCV 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FreeAnchor: Learning to Match Anchors for Visual Object Detection (NeurIPS 2019 · similarity 0.99)
- MetaAnchor: Learning to Detect Objects with Customized Anchors (NeurIPS 2018 · similarity 0.90)
- Probabilistic Anchor Assignment with IoU Prediction for Object Detection (ECCV 2020 · similarity 0.90)
- Multiple Anchor Learning for Visual Object Detection (CVPR 2020 · similarity 0.89)
- Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (CVPR 2020 · similarity 0.89)
- Soft Anchor-Point Object Detection (ECCV 2020 · similarity 0.88)
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.
