Research map / Detection and segmentation / Open-set and few-shot detection
Adapt Foundational Segmentation Models with Heterogeneous Searching Space
Li Yi, Jie Hu, Songan Zhang, Guannan Jiang
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 |
|---|---|---|---|
| augment images with searched heterogeneous policy before zero-shot SAM | SAM ViT-H · NJU2K 10-shot | mIoU: 42.50 -> 80.69 | Improved |
| augment images with searched policy instead of fine-tuning SAM | SAM ViT-H · CAMO 10-shot | mIoU: 50.43 -> 64.34 | Improved |
| augment images with searched policy instead of fine-tuning SAM | SAM ViT-H · Kvasir 10-shot | mIoU: 65.21 -> 71.81 | Improved |
| augment images with searched policy instead of fine-tuning SAM | SAM ViT-H · MTSD 10-shot | mIoU: 59.51 -> 58.33 | Hurt |
| augment images with searched policy instead of fine-tuning SAM | SAM ViT-H · KolektorSDD2 10-shot | mIoU: 55.67 -> 55.83 | No clear effect |
| add radical learning-based methods (depth, normal) to search space (Complete-32 vs Mild-25) (ablation) | policy search · CAMO | mIoU: 51.14 -> 64.34 | Improved |
| add radical methods to search space (Complete-32 vs Mild-25) (ablation) | policy search · NJU2K | mIoU: 52.28 -> 80.69 | Improved |
| add 3 mild learning-based methods (Complete-32 vs Radical-29) (ablation) | policy search · NJU2K | mIoU: 79.61 -> 80.69 | Improved |
| use three-image reward instead of single-image reward (ablation) | policy search · NJU2K | mIoU: 47.95 -> 80.69 | Improved |
| use three-image reward instead of single-image reward (ablation) | policy search · CAMO | mIoU: 50.95 -> 64.34 | Improved |
| use reward over all few-shot images instead of three images (ablation) | policy search · NJU2K | mIoU: 80.69 -> 78.78 | Hurt |
| use reinforcement learning instead of random sampling of 1000 policies (ablation) | policy search · NJU2K | mIoU: 71.63 -> 80.69 | Improved |
Closest papers by meaning
- Rein++: Efficient Generalization and Adaptation for Semantic Segmentation With Vision Foundation Models (TPAMI 2026 · similarity 0.87)
- TrivialAugment: Tuning-Free Yet State-of-the-Art Data Augmentation (ICCV 2021 · similarity 0.85)
- Uncertainty-aware Fine-tuning of Segmentation Foundation Models (NeurIPS 2024 · similarity 0.85)
- Improving the Generalization of Segmentation Foundation Model under Distribution Shift via Weakly Supervised Adaptation (CVPR 2024 · similarity 0.84)
- Adapt Before Comparison: A New Perspective on Cross-Domain Few-Shot Segmentation (CVPR 2024 · similarity 0.84)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · 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.
