Research map / Detection and segmentation / Open-set and few-shot detection
Long-Tailed Instance Segmentation Using Gumbel Optimized Loss
Konstantinos Panagiotis Alexandridis, Jiankang Deng, Anh Nguyen, Shan Luo
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 |
|---|---|---|---|
| replace Sigmoid with Gumbel activation (random sampler) (ablation) | Mask R-CNN R50 · LVIS v1 val | AP: 16.4 -> 19.0 | Improved |
| replace Softmax with Gumbel activation (random sampler) (ablation) | Mask R-CNN R50 · LVIS v1 val | AP rare: 0.0 -> 4.9 | Improved |
| replace Sigmoid with Gumbel activation under RFS sampler (ablation) | Mask R-CNN R50 · LVIS v1 val | AP: 22.0 -> 22.7 | Improved |
| add Gumbel activation to RFS+EQL (ablation) | Mask R-CNN R50 2x · LVIS v1 val | AP: 25.3 -> 26.1 | Improved |
| add stricter NMS 0.3 and Normalised Mask (Enh) (ablation) | RFS+EQL+Gumbel · LVIS v1 val | AP: 26.1 -> 26.9 | Improved |
| replace EQL with DropLoss on top of RFS+Gumbel+Enh (ablation) | GOL · LVIS v1 val | AP: 26.9 -> 27.7 | Improved |
| add Gumbel activation to RFS alone (ablation) | Mask R-CNN R50 2x · LVIS v1 val | AP: 23.7 -> 23.5 | No clear effect |
| swap Sigmoid for Gumbel in EQL (ablation) | EQL · LVIS v1 val | AP: 18.6 -> 21.7 | Improved |
| swap Sigmoid for Gumbel in DropLoss (ablation) | DropLoss · LVIS v1 val | AP: 19.8 -> 22.0 | Improved |
| swap Sigmoid for Gumbel in ACSL (ablation) | ACSL · LVIS v1 val | AP: 20.7 -> 21.0 | Improved |
| swap Sigmoid for Gumbel in Federated Loss (ablation) | Federated Loss · LVIS v1 val | AP: 17.6 -> 20.1 | Improved |
| swap Softmax for Gumbel with Cascade Mask R-CNN R101 (ablation) | Cascade Mask R-CNN R101 · LVIS v1 val | AP: 18.8 -> 22.9 | Improved |
Closest papers by meaning
- Long-Tail Detection with Effective Class-Margins (ECCV 2022 · similarity 0.87)
- DropLoss for Long-Tail Instance Segmentation (AAAI 2021 · similarity 0.86)
- Equalized Focal Loss for Dense Long-Tailed Object Detection (CVPR 2022 · similarity 0.86)
- Equalization Loss v2: A New Gradient Balance Approach for Long-Tailed Object Detection (CVPR 2021 · similarity 0.85)
- SimLTD: Simple Supervised and Semi-Supervised Long-Tailed Object Detection (CVPR 2025 · similarity 0.85)
- Pursuing Better Decision Boundaries for Long-Tailed Object Detection via Category Information Amount (ICLR 2025 · similarity 0.83)
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.
