Research map / Detection and segmentation / CNN features
Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection
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 |
|---|---|---|---|
| Add DGQP based on bounding-box distribution statistics | GFLV1 with ResNet-101 · COCO test-dev | AP: 45.0 -> 46.2 | Helped |
| Add DGQP based on bounding-box distribution statistics (ablation) | GFLV1 with ResNet-50 · COCO minival | AP: 40.2 -> 41.1 | Helped |
| Add DGQP based on bounding-box distribution statistics | RetinaNet, FoveaNet, FCOS, and ATSS · COCO minival | AP: +2.1; +2.1; +2.1; +1.9 | Helped |
| Use distribution statistics instead of convolutional features as DGQP input (ablation) | ATSS with DGQP · COCO minival | AP: 40.8 -> 41.1 | Helped |
| Use decomposed rather than composed classification-IoU representation (ablation) | GFLV2 · COCO minival | AP: 40.7 -> 41.1 | Helped |
| Use Top-4 distribution values together with their mean (ablation) | DGQP · COCO minival | AP: 40.8 -> 41.1 | Helped |
| Add variance to Top-4 values and mean (ablation) | DGQP · COCO minival | AP: 41.1 -> 40.9 | Hurt |
| Use Top-4 instead of Top-2 distribution values (ablation) | DGQP with p=64 · COCO minival | AP: 40.9 -> 41.1 | Helped |
| Increase k from 4 to 8 (ablation) | DGQP with p=64 · COCO minival | AP: 41.1 -> 41.0 | No clear effect |
| Use hidden dimension 64 instead of 32 (ablation) | DGQP with k=4 · COCO minival | AP: 40.9 -> 41.1 | No clear effect |
| Use hidden dimension 64 instead of 128 (ablation) | DGQP with k=4 · COCO minival | AP: 40.9 -> 41.1 | No clear effect |
| Add DGQP to GFLV1 | GFLV1 with ResNet-50 · COCO minival | FPS: 19.4 -> 19.4 | No clear effect |
Closest papers by meaning
- Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection (NeurIPS 2020 · similarity 0.93)
- Pixel-level Quality Assessment for Oriented Object Detection (AAAI 2026 · similarity 0.89)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · similarity 0.85)
- D-FINE: Redefine Regression Task of DETRs as Fine-grained Distribution Refinement (ICLR 2025 · similarity 0.85)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 2024 · similarity 0.85)
- FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 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.