Research map / Detection and segmentation / CNN features
AP-Loss for Accurate One-Stage Object Detection
Kean Chen, Weiyao Lin, Jianguo Li, John See, Ji Wang, Junni Zou
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 CE-loss+OHEM with AP-loss | RetinaNet-ResNet50 · VOC2007 test | AP: 49.1 -> 53.1 | Improved |
| replace Focal loss with AP-loss | RetinaNet-ResNet50 · COCO minival | AP: 33.9 -> 35.0 | Improved |
| replace CE-loss+OHEM with AP-loss | SSD-VGG16 · VOC2007 test | AP: 43.6 -> 45.2 | Improved |
| use AP-loss instead of AUC-loss | RetinaNet-ResNet50 · COCO minival | AP: 25.5 -> 35.0 | Improved |
| use Focal loss instead of CE-loss+OHEM | SSD-VGG16 · VOC2007 test | AP: 43.6 -> 39.3 | Hurt |
| increase batch size from 1 to 8 (ablation) | AP-Loss RetinaNet · VOC2007 test | AP: 52.4 -> 53.1 | Improved |
| set step function delta to 1 rather than 0.25 (ablation) | AP-Loss RetinaNet · VOC2007 test | AP: 50.2 -> 53.1 | Improved |
| set step function delta to 1 rather than 4 (ablation) | AP-Loss RetinaNet · VOC2007 test | AP: 50.5 -> 53.1 | Improved |
| add interpolated AP (ablation) | AP-Loss RetinaNet · VOC2007 test | AP: 52.6 -> 53.1 | Improved |
| add interpolated AP (ablation) | AP-Loss RetinaNet · VOC2007 test | AP75: 57.1 -> 58.1 | Improved |
| use AP-loss instead of Focal loss under Black Patch perturbation | RetinaNet-ResNet50 · VOC2007 test (Black Patch) | mAP50: 60.9 -> 65.2 | Improved |
| use AP-loss instead of Focal loss under Gaussian noise | RetinaNet-ResNet50 · VOC2007 test (Gaussian Noise) | mAP50: 65.7 -> 68.8 | Improved |
Closest papers by meaning
- Towards Accurate One-Stage Object Detection With AP-Loss (CVPR 2019 · similarity 0.99)
- Searching Parameterized AP Loss for Object Detection (NeurIPS 2021 · similarity 0.91)
- DR Loss: Improving Object Detection by Distributional Ranking (CVPR 2020 · similarity 0.90)
- Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection (CVPR 2022 · similarity 0.90)
- Focal Loss for Dense Object Detection (ICCV 2017 · similarity 0.89)
- Bucketed Ranking-based Losses for Efficient Training of Object Detectors (ECCV 2024 · 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.
