Research map / Detection and segmentation / Open-set and few-shot detection
Equalization Loss v2: A New Gradient Balance Approach for Long-Tailed 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 |
|---|---|---|---|
| Use EQL v2 instead of EQL | Mask R-CNN ResNet-50-FPN · LVIS v1.0 val | AP: 21.6 -> 25.5 | Helped |
| Use EQL v2 instead of EQL | Faster R-CNN ResNet-50-FPN · Open Images Challenge 2019 val | AP: 45.3 -> 52.6 | Helped |
| Add the category-agnostic task (ablation) | EQL v2 on Mask R-CNN ResNet-50-FPN · LVIS v1.0 val | AP: 16.1 -> 18.1 | Helped |
| Add negative-gradient reweighting (ablation) | EQL v2 with category-agnostic task · LVIS v1.0 val | AP: 18.1 -> 19.7 | Helped |
| Add positive-gradient reweighting (ablation) | EQL v2 with category-agnostic task and negative-gradient reweighting · LVIS v1.0 val | AP: 19.7 -> 23.7 | Helped |
| Increase alpha from 0 to 8 (ablation) | EQL v2 · LVIS v1.0 val | AP: 19.7 -> 24.0 | Helped |
| Initialize the representation with EQL v2 rather than standard training (ablation) | cRT · LVIS v1.0 val | AP: 22.0 -> 22.4 | No clear effect |
| Initialize the representation with EQL v2 rather than standard training (ablation) | LWS · LVIS v1.0 val | AP: 17.0 -> 23.1 | Helped |
| Initialize the representation with EQL v2 rather than standard training (ablation) | BAGS · LVIS v1.0 val | AP: 23.1 -> 24.0 | Helped |
| Use 2C-softmax rather than C-sigmoid with objectness and negative-gradient reweighting (ablation) | EQL v2 · LVIS v1.0 val | AP: 19.7 -> 20.7 | Helped |
| Use 2C-softmax rather than C-sigmoid with all three components (ablation) | EQL v2 · LVIS v1.0 val | AP: 23.7 -> 23.7 | No clear effect |
| Use the proposed sigmoid-like mapping instead of sqrt mapping, with negative and positive reweighting (ablation) | EQL v2 · LVIS v1.0 val | AP: 21.4 -> 23.7 | Helped |
Closest papers by meaning
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · similarity 0.89)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.89)
- Pursuing Better Decision Boundaries for Long-Tailed Object Detection via Category Information Amount (ICLR 2025 · similarity 0.87)
- Decoupled Contrastive Learning for Long-Tailed Recognition (AAAI 2024 · similarity 0.85)
- Exploring Weight Balancing on Long-Tailed Recognition Problem (ICLR 2024 · similarity 0.85)
- Long-tailed Recognition with Model Rebalancing (NeurIPS 2025 · 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.