Research map / Detection and segmentation / Open-set and few-shot detection
Segment Anything in High Quality
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 |
|---|---|---|---|
| add HQ-Output Token and HQ-Features to SAM | SAM ViT-L · DIS val | mBIoU: 52.8 to 70.4 | Helped |
| add HQ-Output Token and HQ-Features to SAM | SAM ViT-L · COCO zero-shot | AP_B: 33.3 -> 34.4 | Helped |
| HQ-Output Token instead of extra context tokens (ablation) | SAM · Four HQ datasets | mBIoU: 77.0 -> 79.2 | Helped |
| add global-local HQ-Features fusion (ablation) | HQ-SAM · Four HQ datasets | mBIoU: 79.2 -> 81.8 | Helped |
| add early-layer ViT feature (ablation) | HQ-SAM · Four HQ datasets | mBIoU: 81.3 to 81.8 | Helped |
| add final-layer ViT encoder feature (ablation) | HQ-SAM · Four HQ datasets | mBIoU: 80.1 to 81.3 | Helped |
| replace global-local fusion with FPN (ablation) | HQ-SAM · Four HQ datasets | mIoU: 89.1 to 87.4 | Hurt |
| boundary-only loss or dot product with SAM token (ablation) | HQ-Output Token · Four HQ datasets | mBIoU | No clear effect |
| finetune SAM mask decoder instead of token adaptation | SAM · COCO | AP_B: 33.3 -> 9.0 | Hurt |
| tune encoder with adapters | SAM · COCO | AP_B: 33.3 to 29.6 | Hurt |
| train on HQSeg-44K vs 44K SA-1B samples (ablation) | HQ-SAM · Four HQ datasets | mBIoU: 70.1 to 81.8 | Helped |
| replace SAM with HQ-SAM, coarse mask prompt | SAM · BIG test | mBIoU: 41.8 -> 75.1 | Helped |
Closest papers by meaning
- SAMRefiner: Taming Segment Anything Model for Universal Mask Refinement (ICLR 2025 · similarity 0.90)
- BA-SAM: Scalable Bias-Mode Attention Mask for Segment Anything Model (CVPR 2024 · similarity 0.90)
- EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything (CVPR 2024 · similarity 0.90)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.89)
- AoP-SAM: Automation of Prompts for Efficient Segmentation (AAAI 2025 · similarity 0.89)
- Stable Segment Anything Model (ICLR 2025 · similarity 0.89)
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.