Research map / Detection and segmentation / Open-set and few-shot detection
Towards Training-Free Open-World Segmentation via Image Prompt Foundation Models
Lv Tang, Peng-Tao Jiang, Haoke Xiao, Bo Li
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 Stable Diffusion features to DINOv2 features (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 41.7 -> 43.0 | Improved |
| add DINOv2 features to SD features (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 21.9 -> 43.0 | Improved |
| filter prompt background with TSDN (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 9.4 -> 43.0 | Improved |
| use ground-truth mask instead of TSDN to filter prompt (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 43.0 -> 47.3 | Improved |
| replace TSDN with A2S saliency detector (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 43.0 -> 42.4 | No clear effect |
| cluster top-K points into c=4 centers instead of c=32 (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 28.3 -> 43.0 | Improved |
| reduce cluster centers from 4 to 2 (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 43.0 -> 39.1 | Hurt |
| increase K from 4 to 32 (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 40.1 -> 43.0 | Improved |
| increase K from 32 to 64 (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 43.0 -> 40.1 | Hurt |
| use DINOv2 instead of MAE features (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 10.7 -> 43.0 | Improved |
| use DINOv2 instead of CLIP features (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 24.7 -> 43.0 | Improved |
| vary image prompt set (set-1 to set-3) (ablation) | IPSeg · COCO-20i (mean over 4 folds) | mIoU: 41.9 -> 43.0 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SegGPT — SegGPT: Towards Segmenting Everything in Context (ICCV 2023) · better on 0 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts (NeurIPS 2025 · similarity 0.88)
- Just a Few Glances: Open-Set Visual Perception with Image Prompt Paradigm (AAAI 2025 · similarity 0.88)
- Prompting Diffusion Models for Zero-Shot Instance Segmentation (NeurIPS 2026 · similarity 0.88)
- Segment Anyword: Mask Prompt Inversion for Open-Set Grounded Segmentation (ICML 2025 · similarity 0.88)
- APSeg: Auto-Prompt Network for Cross-Domain Few-Shot Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Visual In-Context Prompting (CVPR 2024 · similarity 0.86)
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.
