Research map / Detection and segmentation / Open-set and few-shot detection
Equalization Loss for Long-Tailed Object Recognition
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 |
|---|---|---|---|
| Ignore frequent-category negative gradients for rare categories using EQL | ResNet-50 Mask R-CNN · LVIS v0.5 val | mask AP: 20.1 -> 22.8 | Helped |
| Ignore frequent-category negative gradients for rare categories using EQL | ResNet-50 Mask R-CNN · LVIS v0.5 val | rare-category AP: 7.2 -> 11.3 | Helped |
| Ignore frequent-category negative gradients for rare categories using EQL | ResNet-50 Mask R-CNN · LVIS v0.5 val | frequent-category AP: 25.4 -> 25.1 | No clear effect |
| Use EQL rather than sigmoid cross-entropy | ResNet-50 · OID19 val | AP: 48.13 -> 57.83 | Helped |
| Use SEQL rather than the softmax-loss baseline | ResNet-32 · CIFAR-100-LT test | Acc@top1: 41.33 -> 43.38 | Helped |
| Use SEQL rather than the softmax-loss baseline | ResNet-10 · ImageNet-LT test | Acc@top1: 34.7 -> 36.4 | Helped |
| Set λ to include rare and common categories rather than λ=0 (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | mask AP: 20.1 -> 22.8 | Helped |
| Set λ to include rare and common categories rather than λ=0 (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | frequent-category AP: 25.4 -> 25.1 | No clear effect |
| Replace threshold function with exponential decay (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | mask AP: 22.8 -> 22.3 | Hurt |
| Replace threshold function with Gompertz decay (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | mask AP: 22.8 -> 22.7 | No clear effect |
| Add E(r) to restrict equalization to foreground proposals (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | mask AP: 22.2 -> 22.8 | Helped |
| Add E(r) to retain background-proposal gradients (ablation) | EQL on ResNet-50 Mask R-CNN · LVIS v0.5 val | frequent-category AP: 23.1 -> 25.1 | Helped |
Closest papers by meaning
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.89)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.87)
- OFBD: Object-Focused Background Debiasing for Long-Tailed Learning (NeurIPS 2026 · similarity 0.86)
- Pursuing Better Decision Boundaries for Long-Tailed Object Detection via Category Information Amount (ICLR 2025 · similarity 0.86)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.86)
- Search and Detect: Training-Free Long Tail Object Detection via Web-Image Retrieval (CVPR 2025 · 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.