Research map / Detection and segmentation / Open-set and few-shot detection
The Equalization Losses: Gradient-Driven Training for Long-tailed Object Recognition
Jingru Tan, Bo Li, Xin Lu, Yongqiang Yao, Fengwei Yu, Tong He, Wanli Ouyang
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 objectness branch (ablation) | Sigmoid-EQL on Mask R-CNN · LVIS v1 val (1x) | AP: 16.1 -> 18.1 | Improved |
| add negative gradient reweighting (ablation) | Sigmoid-EQL (with obj) · LVIS v1 val (1x) | AP: 18.1 -> 19.7 | Improved |
| add positive gradient reweighting (ablation) | Sigmoid-EQL (with obj) · LVIS v1 val (1x) | AP: 18.1 -> 23.3 | Improved |
| add negative reweighting on top of obj+pos (ablation) | Sigmoid-EQL · LVIS v1 val (1x) | AP: 23.3 -> 23.7 | Improved |
| use sigmoid-like mapping instead of exp mapping (ablation) | Sigmoid-EQL · LVIS v1 val (neg+pos) | AP: 23.2 -> 23.7 | Improved |
| use sigmoid-like mapping instead of sqrt mapping (ablation) | Sigmoid-EQL · LVIS v1 val (neg+pos) | AP_r: 6.2 -> 14.9 | Improved |
| add Sigmoid-EQL to CE baseline, 3x schedule | Mask R-CNN R50 · LVIS v1 val (3x) | AP: 20.5 -> 26.2 | Improved |
| add focusing factor (ablation) | EFL baseline · LVIS v1 val | AP: 25.7 -> 26.2 | Improved |
| add weighting factor (ablation) | EFL baseline · LVIS v1 val | AP: 25.7 -> 26.1 | Improved |
| add both factors (EFL) to improved baseline | ATSS baseline · LVIS v1 val R50 | AP: 25.7 -> 27.5 | Improved |
| raise scale s from 8 to 12 (ablation) | EFL · LVIS v1 val | AP: 27.5 -> 26.5 | Hurt |
| train EFL with ResNet-101 instead of ResNet-50 | EFL · LVIS v1 val | AP_r: 20.2 -> 23.5 | 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
- Equalization Loss v2: A New Gradient Balance Approach for Long-Tailed Object Detection (CVPR 2021 · similarity 0.94)
- Equalization Loss for Long-Tailed Object Recognition (CVPR 2020 · similarity 0.94)
- Equalized Focal Loss for Dense Long-Tailed Object Detection (CVPR 2022 · similarity 0.91)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.90)
- Long-Tailed Recognition via Weight Balancing (CVPR 2022 · similarity 0.90)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.90)
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.
