Research map / Detection and segmentation / Open-set and few-shot detection
SHERPA: Fine-tuning Segment Anything Models with Task-relevant Guidance
Jingcheng Xie, Yinda Chen, Xiaoyu Liu, Haoyuan Shi, Zhiwei Xiong
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 FRS module to small-model guidance (ablation) | SHERPA · DISK-5K | F1 general: 0.5752 -> 0.7930 | Improved |
| add FRS module to small-model guidance (ablation) | SHERPA · DUTS | F1 general: 0.6432 -> 0.8284 | Improved |
| add FRS module to small-model guidance (ablation) | SHERPA · Lucchi | F1 general: 0.7611 -> 0.7857 | Improved |
| add FRS module to small-model guidance (ablation) | SHERPA · DISK-5K | F1 valid: 0.8695 -> 0.8855 | Improved |
| use small SAM instead of large SAM itself as guide (ablation) | SHERPA · DISK-5K | F1 general: 0.7797 -> 0.7930 | Improved |
| fine-tune small guiding model with KL constraint (ablation) | SHERPA · DISK-5K | F1 general: 0.7812 -> 0.7930 | Improved |
| use task-relevant feature guidance instead of regularizing residual components (ablation) | SHERPA · DISK-5K | F1 general: 0.7431 -> 0.7930 | Improved |
| set Var Ratio to 0.75 instead of 1.00 (all channels) (ablation) | SHERPA · DISK-5K | F1 general: 0.5257 -> 0.7930 | Improved |
| set Var Ratio to 0.75 instead of 0.00 (ablation) | SHERPA · DISK-5K | F1 general: 0.7234 -> 0.7930 | Improved |
| use SHERPA instead of standard fine-tuning | SAM ViT-H · DISK-5k generalization retention | retention: 85.3% -> 93.4% | Improved |
| use SHERPA instead of standard fine-tuning | SAM ViT-H · Lucchi generalization retention | retention: 78.9% -> 91.8% | Improved |
| use SHERPA instead of standard fine-tuning | SAM2 · VOST generalization retention | retention: 84.0% -> 95.1% | Improved |
Closest papers by meaning
- Uncertainty-aware Fine-tuning of Segmentation Foundation Models (NeurIPS 2024 · similarity 0.89)
- InfoSAM: Fine-Tuning the Segment Anything Model from An Information-Theoretic Perspective (ICML 2025 · similarity 0.88)
- SAM-PARSER: Fine-Tuning SAM Efficiently by Parameter Space Reconstruction (AAAI 2024 · similarity 0.87)
- Personalize Segment Anything Model with One Shot (ICLR 2024 · similarity 0.85)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.85)
- TinySAM: Pushing the Envelope for Efficient Segment Anything Model (AAAI 2025 · similarity 0.85)
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.
