Research map / Detection and segmentation / Open-set and few-shot detection
Mask Matching Transformer for Few-Shot Segmentation
Read the paperCheck 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 heuristic top-similarity selection with learnable matching that fuses proposals (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 42.1 -> 56.4; 22.4 -> 37.9 | Helped |
| Add Self-Alignment to heuristic matching (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 42.1 -> 46.4; 22.4 -> 23.6 | Helped |
| Use non-parametric rather than learned Cross-Alignment (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 56.1 -> 38.0; 35.3 -> 16.7 | Hurt |
| Add learned Cross-Alignment (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 42.1 -> 56.1; 22.4 -> 35.3 | Helped |
| Add cross-alignment supervision L_co (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 62.1 -> 63.3; 43.2 -> 44.2 | Helped |
| Use features after MSDeformAttn instead of backbone features for matching (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 63.3 -> 60.5; 44.2 -> 42.8 | Hurt |
| Train POS and MM end-to-end rather than in two stages (ablation) | MM-Former · Pascal-5^i and COCO-20^i, 1-shot | mean IoU: 63.3 -> 60.6; 44.2 -> 40.3 | Hurt |
| Remove the mask classifier from POS (ablation) | MM-Former POS · Pascal-5^i and COCO-20^i, POS oracle | mean IoU: 78.9 -> 82.5; 68.4 -> 68.9 | Helped |
| Increase mask proposals from 10 to 50 (ablation) | MM-Former POS · Pascal-5^i, 1-shot | Final mean IoU: 57.9 -> 61.0 | Helped |
| Increase mask proposals from 50 to 100 (ablation) | MM-Former POS · Pascal-5^i, 1-shot | Final mean IoU: 61.0 -> 63.3 | Helped |
| Train on more images from only the 15 Pascal training classes (COCO-15) (ablation) | MM-Former · Pascal-5^i transfer, 1-shot and 5-shot | mIoU: 63.3 -> 60.7; 64.9 -> 64.8 | Hurt |
| Train on more categories using a similarly sized sample (COCO-75-sub) (ablation) | MM-Former · Pascal-5^i transfer, 1-shot and 5-shot | mIoU: 63.3 -> 66.8; 64.9 -> 68.9 | Helped |
Closest papers by meaning
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.90)
- Foreground-Covering Prototype Generation and Matching for SAM-Aided Few-Shot Segmentation (AAAI 2025 · similarity 0.88)
- Unlocking the Potential of Pre-trained Vision Transformers for Few-Shot Semantic Segmentation through Relationship Descriptors (CVPR 2024 · similarity 0.87)
- PEM: Prototype-based Efficient MaskFormer for Image Segmentation (CVPR 2024 · similarity 0.86)
- Lightweight Frequency Masker for Cross-Domain Few-Shot Semantic Segmentation (NeurIPS 2024 · similarity 0.86)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · 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.