Research map / Detection and segmentation / Open-set and few-shot detection
Segment Anything
Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dollar, Ross Girshick
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 |
|---|---|---|---|
| predict multiple output masks (ambiguity-aware) instead of a single mask (ablation) | SAM · LVIS v1 object proposals | mask AR@1000: 54.9 -> 59.3 | Improved |
| use ambiguity-aware multi-mask output (ablation) | SAM · 23 datasets single-point human study | mask quality rating | Improved |
| train on automatic masks only instead of all three data-engine stages (ablation) | SAM · 23-dataset suite | mIoU: ~0.5 mIoU lower | No clear effect |
| add each data engine stage cumulatively to training data (ablation) | SAM · 23-dataset suite | mIoU | Improved |
| oversample manual and semi-automatic masks by 10x when training with all stages (ablation) | SAM · 23-dataset suite | mIoU | Improved |
| train on 1M images (about 10% of SA-1B) instead of all 11M (ablation) | SAM · 23-dataset suite | mIoU | No clear effect |
| reduce training data to 0.1M images (ablation) | SAM · 23-dataset suite | mIoU | Hurt |
| scale image encoder from ViT-B to ViT-H (ablation) | SAM · 23-dataset suite | mIoU | Improved |
| scale image encoder from ViT-L to ViT-H (ablation) | SAM · 23-dataset suite | mIoU | No clear effect |
| SAM vs RITM on single-point prompts (mIoU, 23 datasets) | SAM · 23-dataset suite | mIoU: higher on 16 of 23 datasets | Improved |
| use zero-shot SAM for instance segmentation instead of fully supervised ViTDet masks | SAM with ViTDet boxes · COCO | mask AP: 51.0 -> 46.5 | Hurt |
| add more point prompts (1 to 9) to interactive segmentation | SAM vs baselines · 23-dataset suite | mIoU | Mixed |
Closest papers by meaning
- TinySAM: Pushing the Envelope for Efficient Segment Anything Model (AAAI 2025 · similarity 0.91)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.91)
- EfficientSAM: Leveraged Masked Image Pretraining for Efficient Segment Anything (CVPR 2024 · similarity 0.91)
- SAM 2: Segment Anything in Images and Videos (ICLR 2025 · similarity 0.91)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · similarity 0.89)
- RobustSAM: Segment Anything Robustly on Degraded Images (CVPR 2024 · 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.
