Research map / Detection and segmentation / CNN features
Towards Accurate One-Stage Object Detection With AP-Loss
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 Focal Loss with AP-Loss | RetinaNet · VOC2007 test | AP: 51.3 -> 53.1 | Helped |
| Replace Focal Loss with AP-Loss | RetinaNet · COCO minival | AP: 33.9 -> 35.0 | Helped |
| Replace Focal Loss with AP-Loss | SSD · VOC2007 test | AP: 39.3 -> 45.2 | Helped |
| Replace RetinaNet's Focal Loss with AP-Loss | RetinaNet500, ResNet-101 · COCO test-dev | AP: 34.4 -> 37.4 | Helped |
| Use batch size 8 rather than batch size 1 (ablation) | RetinaNet with AP-Loss · VOC2007 test | AP: 52.4 -> 53.1 | Helped |
| Set piecewise-step δ to 1 rather than 0.25 (ablation) | RetinaNet with AP-Loss · VOC2007 test | AP: 50.2 -> 53.1 | Helped |
| Set piecewise-step δ to 1 rather than 0.5 (ablation) | RetinaNet with AP-Loss · VOC2007 test | AP: 51.3 -> 53.1 | Helped |
| Set piecewise-step δ to 2 rather than 1 (ablation) | RetinaNet with AP-Loss · VOC2007 test | AP: 53.1 -> 52.8 | Hurt |
| Add interpolation to AP (ablation) | RetinaNet with AP-Loss · VOC2007 test | AP: 52.6 -> 53.1 | Helped |
| Replace AP-Loss with AUC-Loss | RetinaNet · COCO minival | AP: 35.0 -> 25.5 | Hurt |
| Optimize AP-Loss with approximate gradient descent instead of error-driven updates | RetinaNet · VOC2007 trainval | AP-loss convergence | Hurt |
| Optimize AP-Loss with structured hinge loss instead of error-driven updates | RetinaNet · VOC2007 trainval | AP-loss convergence | Hurt |
Closest papers by meaning
- Searching Parameterized AP Loss for Object Detection (NeurIPS 2021 · similarity 0.90)
- DR Loss: Improving Object Detection by Distributional Ranking (CVPR 2020 · similarity 0.89)
- Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection (CVPR 2022 · similarity 0.88)
- RankDetNet: Delving Into Ranking Constraints for Object Detection (CVPR 2021 · similarity 0.87)
- A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection (NeurIPS 2020 · similarity 0.86)
- Equalized Focal Loss for Dense Long-Tailed Object Detection (CVPR 2022 · similarity 0.86)
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.