Research map / Detection and segmentation / Weakly supervised detection and segmentation
WPS-SAM: Towards Weakly-Supervised Part Segmentation with Foundation Models
Xin-Jian Wu, Ruisong Zhang, Jie Qin, Shijie Ma, Cheng-Lin Liu
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 |
|---|---|---|---|
| learn part prompts end-to-end from SAM features with a distilled prompter, using only box labels | fully supervised part segmentation models · PartImageNet val | mIoU (%): 65.24 -> 68.93 | Improved |
| replace Det-SAM detector pipeline with learned query-based prompter (ablation) | Det-SAM (Faster-RCNN ResNet-50) · PartImageNet val | mIoU (%): 63.7 -> 68.9 | Improved |
| use SAM encoder as detector backbone in Det-SAM (ablation) | Det-SAM · PartImageNet val | mIoU (%): 63.7 -> 54.6 (Faster-RCNN) | Hurt |
| use direct prompter instead of SAM pseudo-label paradigm for training MaskFormer (ablation) | MaskFormer Swin-B · PartImageNet val | mIoU (%): 56.83 -> 68.93 | Improved |
| use bounding box labels instead of center point labels (ablation) | WPS-SAM · PartImageNet val | mIoU (%) | Improved |
| increase number of queries from 25 to 50/75/100 (ablation) | WPS-SAM prompter · PartImageNet val | mIoU (%) | Hurt |
| fine-tune the SAM encoder instead of freezing it (ablation) | WPS-SAM · PartImageNet val | mIoU (%): 68.93 -> 65.16 | Hurt |
| use query-based transformer prompter instead of anchor-based CNN (ablation) | student prompter · PartImageNet val | mIoU (%): 57.23 -> 68.93 | Improved |
| increase beta (matching/regression weight) from 1 to 20 with alpha 5, lambda_cls 10 (ablation) | WPS-SAM losses · PartImageNet val | mIoU (%): 61.93 -> 68.93 | Improved |
| increase beta to 50 (ablation) | WPS-SAM losses · PartImageNet val | mIoU (%): 68.93 -> 67.82 | Hurt |
Closest papers by meaning
- From SAM to CAMs: Exploring Segment Anything Model for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.90)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.88)
- AoP-SAM: Automation of Prompts for Efficient Segmentation (AAAI 2025 · similarity 0.88)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.88)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 2024 · similarity 0.87)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.87)
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.
