Research map / Detection and segmentation / Open-set and few-shot detection
Adaptive Class Suppression Loss for Long-Tail 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 |
|---|---|---|---|
| Replace the classification loss with ACSL | Faster R-CNN ResNet-50-FPN · LVIS-v0.5 val | mAP: 21.18 -> 26.36 | Helped |
| Replace the classification loss with ACSL | ResNet50-FPN detector · Open Images validation | AP: 55.1 -> 60.3 | Helped |
| Use ACSL with ResNet101-FPN | Open Images detector · Open Images validation | AP: 56.3 -> 61.6 | Helped |
| Use ACSL with ResNet152-FPN | Open Images detector · Open Images validation | AP: 57.4 -> 62.8 | Helped |
| Set BAGS grouping cutoff to 5 rather than the more suitable intermediate range (ablation) | BAGS · LVIS | mAP: 22.74 | Hurt |
| Set BAGS grouping cutoff to 5000 rather than the more suitable intermediate range (ablation) | BAGS · LVIS | mAP: 23.89 | Hurt |
| Train the baseline for 24 rather than 12 epochs (ablation) | Faster R-CNN ResNet50-FPN baseline · LVIS | mAP: 21.18 -> 22.28 | Helped |
| Set ACSL suppression threshold ξ to 0.7 (ablation) | Faster R-CNN ResNet50-FPN · LVIS | mAP: 22.28 -> 26.36 | Helped |
| Increase ACSL threshold ξ beyond 0.7 to 0.9 (ablation) | Faster R-CNN ResNet50-FPN · LVIS | mAP: 26.36 -> 25.99 | Hurt |
| Apply ACSL to the classifier | Faster R-CNN ResNet-101 · LVIS-v0.5 val | mAP: 22.36 -> 27.49 | Helped |
| Apply ACSL to the classifier | Faster R-CNN ResNeXt-101-64x4d · LVIS-v0.5 val | mAP: 24.70 -> 28.93 | Helped |
| Use ACSL instead of the baseline loss | Cascade R-CNN ResNet-101 · LVIS-v0.5 val | mAP: 25.14 -> 29.71 | Helped |
Closest papers by meaning
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.88)
- Pursuing Better Decision Boundaries for Long-Tailed Object Detection via Category Information Amount (ICLR 2025 · similarity 0.87)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.87)
- EAT: Towards Long-Tailed Out-of-Distribution Detection (AAAI 2024 · similarity 0.86)
- CQ-DINO: Mitigating Gradient Dilution via Category Queries for Vast Vocabulary Object Detection (NeurIPS 2025 · similarity 0.86)
- OFBD: Object-Focused Background Debiasing for Long-Tailed Learning (NeurIPS 2026 · 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.