Research map / Detection and segmentation / Open-set and few-shot detection
MultiSeg: Semantically Meaningful, Scale-Diverse Segmentations From Minimal User Input
Jun Hao Liew, Scott Cohen, Brian Price, Long Mai, Sim-Heng Ong, Jiashi Feng
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 scale-prior branches (scale diversity) instead of ranked diversity loss (ablation) | MultiSeg · Fashionista (object parts) | #Total interactions: 12.01 -> 8.61 | Improved |
| add scale priors instead of ranked diversity loss (fewer proposal selections) (ablation) | MultiSeg · Fashionista (object parts) | #Select: 3.87 -> 0.96 | Improved |
| add scale priors instead of ranked diversity loss in the user study (ablation) | MultiSeg · User study (50 images, 5 participants) | #Total interactions: 4.38 -> 3.54 | Improved |
| add scale priors instead of ranked diversity loss in the user study (time) (ablation) | MultiSeg · User study (50 images, 5 participants) | Time (s): 9.47 -> 7.55 | Improved |
| use MultiSeg instead of single-output DIOS in the user study | DeepLabv3+ interactive segmentation · User study (50 images, 5 participants) | Time (s): 8.84 -> 7.55 | Improved |
| use MultiSeg instead of single-output DIOS on object parts | DeepLabv3+ interactive segmentation · Fashionista (object parts) | #Total interactions: 9.13 -> 8.61 | Improved |
| retrain DIOS with the same DeepLabv3+ backbone | DIOS · PASCAL VOC val (85% IoU) | clicks to 85% IoU: 6.88 -> 3.51 | Improved |
| use MultiSeg instead of backbone-matched DIOS for single objects | DeepLabv3+ interactive segmentation · PASCAL VOC val (85% IoU) | clicks to 85% IoU: 3.51 -> 3.88 | No clear effect |
| use MultiSeg instead of backbone-matched DIOS on Berkeley | DeepLabv3+ interactive segmentation · Berkeley (90% IoU) | clicks to 90% IoU: 4.31 -> 4.00 | Improved |
| add synthesized diverse training samples (versus MultiSeg-SingleGt) (ablation) | MultiSeg · PASCAL VOC val (m-diversity score) | m-diversity score | Improved |
| define output branches by scale instead of ranked diversity loss (ablation) | MultiSeg · PASCAL VOC val (m-diversity score) | m-diversity score | Improved |
Closest papers by meaning
- Rethinking Interactive Image Segmentation with Low Latency High Quality and Diverse Prompts (CVPR 2024 · similarity 0.89)
- MultiverSeg: Scalable Interactive Segmentation of Biomedical Imaging Datasets with In-Context Guidance (ICCV 2025 · similarity 0.89)
- Interactive Image Segmentation With Latent Diversity (CVPR 2018 · similarity 0.89)
- Scaling up Image Segmentation across Data and Tasks (CVPR 2025 · similarity 0.88)
- Prompting Diffusion Models for Zero-Shot Instance Segmentation (NeurIPS 2026 · similarity 0.88)
- OMG-Seg: Is One Model Good Enough For All Segmentation? (CVPR 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.
