Research map / Detection and segmentation / Open-set and few-shot detection
Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax
Yu Li, Tao Wang, Bingyi Kang, Sheng Tang, Chunfeng Wang, Jintao Li, Jiashi Feng
Check 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 the standard classifier with balanced group softmax | Faster R-CNN R50-FPN · LVIS val | mAP: 20.98 -> 25.96 | Improved |
| Replace the standard classifier with balanced group softmax | Faster R-CNN R50-FPN · LVIS val | AP₁: 0.00 -> 11.33 | Improved |
| Use BAGS | Faster R-CNN R50-FPN · LVIS val | mAP: 20.98 -> 25.96 | Improved |
| Use BAGS | Cascade R-CNN X101-FPN · LVIS val | mAP: 27.16 -> 32.77 | Improved |
| Use BAGS | Mask R-CNN R50-FPN · LVIS val | mask mAP: 20.68 -> 26.25 | Improved |
| Use BAGS | Faster R-CNN R50-FPN · COCO-LT val | mAP: 20.3 -> 22.5 | Improved |
| Add a separate background group to group softmax (ablation) | BAGS with four normal-category groups · LVIS val | mAP: 17.82 -> 19.73 | Improved |
| Add an others class to every group (ablation) | Group softmax with four groups · LVIS val | mAP: +2.7% | Improved |
| Add both the background group and others classes (ablation) | Group softmax with four groups · LVIS val | mAP: 23.74 -> 25.96 | Improved |
| Use 2 rather than 4 normal-category groups (ablation) | BAGS · LVIS val | mAP: 25.96 -> 25.31 | Hurt |
| Use 8 rather than 4 normal-category groups (ablation) | BAGS · LVIS val | mAP: 25.96 -> 24.85 | Hurt |
| Increase the others sampling ratio up to β=8 (ablation) | BAGS · LVIS val | mAP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Faster R-CNN — Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks (TPAMI 2016) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adaptive Class Suppression Loss for Long-Tail Object Detection (CVPR 2021 · similarity 0.89)
- The Equalization Losses: Gradient-Driven Training for Long-tailed Object Recognition (TPAMI 2023 · similarity 0.89)
- Large-Scale Object Detection in the Wild With Imbalanced Data Distribution, and Multi-Labels (TPAMI 2024 · similarity 0.89)
- Deep Long-Tailed Learning: A Survey (TPAMI 2023 · similarity 0.89)
- Exploring Classification Equilibrium in Long-Tailed Object Detection (ICCV 2021 · similarity 0.89)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.87)
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.
