Research map / Label-efficient and robust learning / Recognition and long-tail
Seesaw Loss for Long-Tailed Instance Segmentation
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 |
|---|---|---|---|
| Replace Cross-Entropy with Seesaw Loss plus normalized mask prediction | Mask R-CNN · LVIS v1 val, random sampler | AP: 20.6 -> 27.1 | Helped |
| Replace Cross-Entropy with Seesaw Loss plus normalized mask prediction | Mask R-CNN · LVIS v1 val, RFS | AP: 25.5 -> 28.1 | Helped |
| Replace Cross-Entropy with Seesaw Loss plus normalized mask prediction | Cascade Mask R-CNN · LVIS v1 val, random sampler | AP: 22.6 -> 29.6 | Helped |
| Replace Cross-Entropy with Seesaw Loss plus normalized mask prediction | Cascade Mask R-CNN · LVIS v1 val, RFS | AP: 27.0 -> 30.1 | Helped |
| Add normalized mask prediction to Seesaw Loss (ablation) | Mask R-CNN · LVIS v1 val, random sampler | AP: 26.6 -> 27.1 | Helped |
| Add normalized mask prediction to Seesaw Loss (ablation) | Mask R-CNN · LVIS v1 val, RFS | AP: 27.6 -> 28.1 | Helped |
| Add the mitigation factor (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 23.7 -> 25.1 | Helped |
| Add the compensation factor (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 23.7 -> 24.1 | Helped |
| Add compensation alongside mitigation (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 25.1 -> 25.7 | Helped |
| Add normalized linear activation (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 23.7 -> 24.7 | Helped |
| Use online class-count accumulation instead of dataset frequencies (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 26.1 -> 26.4 | Helped |
| Add an objectness branch (ablation) | Seesaw, Mask R-CNN · LVIS v1 val, RFS | AP: 25.3 -> 26.4 | Helped |
Closest papers by meaning
- Long-Tailed Learning as Multi-Objective Optimization (AAAI 2024 · similarity 0.87)
- Pursuing Better Decision Boundaries for Long-Tailed Object Detection via Category Information Amount (ICLR 2025 · similarity 0.87)
- Learning from Neighbors: Category Extrapolation for Long-Tail Learning (CVPR 2025 · similarity 0.85)
- Learning Only When It Matters: Cost-Aware Long-Tailed Classification (AAAI 2024 · similarity 0.85)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · similarity 0.85)
- LT-Soups: Bridging Head and Tail Classes via Subsampled Model Soups (NeurIPS 2025 · similarity 0.84)
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.