Research map / Detection and segmentation / Open-set and few-shot detection
Exploring Classification Equilibrium in Long-Tailed Object Detection
Chengjian Feng, Yujie Zhong, Weilin Huang
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 |
|---|---|---|---|
| add Equilibrium Loss (EBL) to the baseline (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP: 20.2 -> 23.8 | Improved |
| add Memory-augmented Feature Sampling (MFS) to the baseline (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP: 20.2 -> 25.7 | Improved |
| add MFS to the baseline (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP_r: 2.9 -> 18.4 | Improved |
| add MFS to the baseline (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP_f: 30.1 -> 29.8 | Hurt |
| add EBL on top of MFS (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP: 25.7 -> 26.6 | Improved |
| add EBL on top of MFS (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP_f: 29.8 -> 30.7 | Improved |
| add EBL to the baseline (ablation) | Mask R-CNN R-50-FPN · LVIS v1.0 val (segmentation, ablation) | AP_c: 18.2 -> 23.7 | Improved |
| guide EBL and MFS by mean classification score instead of class frequency P(y) (ablation) | LOCE · LVIS v1.0 val (segmentation, ablation) | AP: 23.1 -> 26.6 | Improved |
| guide by mean classification score instead of class frequency (ablation) | LOCE · LVIS v1.0 val (segmentation, ablation) | AP_f: 24.1 -> 30.7 | Improved |
| smoothing coefficient alpha 0.8 -> 0.9 (ablation) | LOCE · LVIS v1.0 val (segmentation, ablation) | AP_r: 17.2 -> 18.5 | No clear effect |
| set background score to 0.01 instead of 0.1 (ablation) | LOCE · LVIS v1.0 val (segmentation, ablation) | AP: 26.0 -> 26.6 | Improved |
| set background score 0.001 instead of 0.01 (ablation) | LOCE · LVIS v1.0 val (segmentation, ablation) | AP: 26.6 -> 26.5 | No clear effect |
Closest papers by meaning
- Boosting Long-tailed Object Detection via Step-wise Learning on Smooth-tail Data (ICCV 2023 · similarity 0.89)
- C2AM Loss: Chasing a Better Decision Boundary for Long-Tail Object Detection (CVPR 2022 · similarity 0.89)
- Overcoming Classifier Imbalance for Long-Tail Object Detection With Balanced Group Softmax (CVPR 2020 · similarity 0.89)
- Equalized Focal Loss for Dense Long-Tailed Object Detection (CVPR 2022 · similarity 0.88)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.86)
- Exploring Weight Balancing on Long-Tailed Recognition Problem (ICLR 2024 · 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.
