Research map / Label-efficient and robust learning / Domain adaptation and generalization
Selective, Regularized, and Calibrated: Harnessing Vision Foundation Models for Cross-Domain Few-Shot Semantic Segmentation
Junyuan Ma, Xunzhi Xiang, Wenbin Li, Qi Fan, Yang Gao
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 Hierarchical Layer Selection (HLS) (ablation) | SSP baseline · CD-FSS avg 5-shot (DINOv3) | mIoU: 63.1 -> 76.7 | Improved |
| add Prior Guided Regularization on top of HLS (ablation) | SSP + HLS · CD-FSS avg 5-shot (DINOv3) | mIoU: 76.7 -> 77.3 | Improved |
| add Pixelwise Adaptive Calibration on top of HLS (ablation) | SSP + HLS · CD-FSS avg 5-shot (DINOv3) | mIoU: 76.7 -> 77.2 | Improved |
| add both PGR and PAC on top of HLS (ablation) | SSP + HLS · CD-FSS avg 5-shot (DINOv3) | mIoU: 76.7 -> 77.9 | Improved |
| use ETR-based HLS instead of Static-Max heuristic selector (ablation) | HERA layer routing · CD-FSS avg 5-shot (DINOv3) | mIoU: 71.9 -> 76.7 | Improved |
| use ETR-based HLS instead of gradient-norm selector (ablation) | HERA layer routing · CD-FSS avg 5-shot (DINOv3) | mIoU: 73.1 -> 76.7 | Improved |
| use ETR-based HLS instead of Static-Max on DINOv2 (ablation) | HERA layer routing · CD-FSS avg 5-shot (DINOv2) | mIoU: 64.0 -> 72.6 | Improved |
| make MLP at routed layer trainable at test time rather than frozen (ablation) | HERA · CD-FSS avg 5-shot (DINOv3) | mIoU: 75.78 -> 77.91 | Improved |
| add all three PAC branches (sim, attn, img) versus a single branch (ablation) | HLS+PGR · CD-FSS avg 5-shot (DINOv3) | mIoU: 77.27 -> 77.91 | Improved |
| include L23 in local fusion candidates with tau=2.0 (ablation) | HERA local fusion · CD-FSS avg 5-shot (DINOv3) | mIoU: 75.49 -> 77.85 | Improved |
| keep PAC gate automatic (T=2) rather than off in 5-shot (ablation) | HERA PAC · CD-FSS avg 5-shot (DINOv3) | mIoU: 76.67 -> 77.91 | Improved |
| use PAC always on versus automatic gate in 1-shot (ablation) | HERA PAC · CD-FSS avg 1-shot (DINOv3) | mIoU: 68.02 -> 68.29 | Improved |
Closest papers by meaning
- The Devil is in Low-Level Features for Cross-Domain Few-Shot Segmentation (CVPR 2025 · similarity 0.90)
- Adapt Before Comparison: A New Perspective on Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.90)
- Adapter Naturally Serves as Decoupler for Cross-Domain Few-Shot Semantic Segmentation (ICML 2025 · similarity 0.90)
- Lightweight Frequency Masker for Cross-Domain Few-Shot Semantic Segmentation (NeurIPS 2024 · similarity 0.89)
- UniDSeg: Unified Cross-Domain 3D Semantic Segmentation via Visual Foundation Models Prior (NeurIPS 2024 · similarity 0.88)
- Unleashing the Power of Visual Foundation Models for Generalizable Semantic Segmentation (AAAI 2025 · similarity 0.88)
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.
