Research map / Detection and segmentation / CNN features
Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection
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 IoU assignment with ATSS adaptive sample selection | RetinaNet (#A=1) · MS COCO minival | AP: 37.0 -> 39.3; +2.3% | Helped |
| Apply ATSS candidate selection while retaining fixed FCOS scale ranges (center sampling) (ablation) | FCOS · MS COCO minival | AP: 37.8 -> 38.6 | Helped |
| Replace FCOS fixed scale-range selection with ATSS adaptive selection (ablation) | FCOS · MS COCO minival | AP: 38.6 -> 39.2 | Helped |
| Use spatial and scale constraints instead of IoU-based sample selection (ablation) | RetinaNet (#A=1) · MS COCO minival | AP: 37.0 -> 37.8 | Helped |
| Use IoU-based sample selection instead of spatial and scale constraints (ablation) | FCOS · MS COCO minival | AP: 37.8 -> 36.9 | Hurt |
| Regress from an anchor point rather than an anchor box, holding sample selection fixed (ablation) | RetinaNet and FCOS · MS COCO minival | AP: 37.0 vs. 36.9; 37.8 vs. 37.8 | No clear effect |
| Add GroupNorm, GIoU loss, in-box positives, centerness, and a trainable scalar (ablation) | RetinaNet (#A=1) · MS COCO minival | AP: 32.5 -> 37.0 | Helped |
| Vary ATSS candidate count k from 7 to 17 (ablation) | ATSS · MS COCO minival | AP: 39.1 to 39.3 | No clear effect |
| Reduce candidate count to k=3 (ablation) | ATSS · MS COCO minival | AP: 39.3 -> 38.0 | Hurt |
| Increase candidate count to k=19 (ablation) | ATSS · MS COCO minival | AP: 39.3 -> 38.9 | Hurt |
| Vary anchor scale from 5 to 9 or aspect ratio from 1:4 to 4:1 (ablation) | ATSS · MS COCO minival | AP: AP values from 38.9 to 39.3 | No clear effect |
| Use multiple anchors per location instead of one under ATSS (ablation) | RetinaNet with ATSS · MS COCO minival | AP: 39.2 vs. 39.3 | No clear effect |
Closest papers by meaning
- Feature Selective Anchor-Free Module for Single-Shot Object Detection (CVPR 2019 · similarity 0.90)
- Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (AAAI 2021 · similarity 0.89)
- Unbiased Teacher v2: Semi-Supervised Object Detection for Anchor-Free and Anchor-Based Detectors (CVPR 2022 · similarity 0.89)
- Shape-Adaptive Selection and Measurement for Oriented Object Detection (AAAI 2022 · similarity 0.89)
- Multiple Anchor Learning for Visual Object Detection (CVPR 2020 · similarity 0.89)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (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.