Research map / Detection and segmentation / Open-set and few-shot detection
Predicting Foreground Object Ambiguity and Efficiently Crowdsourcing the Segmentation(s)
Danna Gurari, Kun He, Bo Xiong, Jianming Zhang, Mehrnoosh Sameki, Suyog Dutt Jain, Stan Sclaroff, Margrit Betke, Kristen Lorraine Grauman
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 |
|---|---|---|---|
| fine-tune the SOS network on STATIC ambiguity labels instead of using SOS as is | ambiguity classifier · seven CV benchmarks | average precision: 85.7% -> 90.7% | Improved |
| use CNN-FT instead of Feng et al. saliency detector | ambiguity classifier · VizWiz images | average precision: 54.2% -> 64.6% | Improved |
| apply the classifier to blind-photographer VizWiz images instead of curated benchmark images | CNN-FT · ambiguity prediction AP | average precision: 90.7% -> 64.6% | Hurt |
| ask crowd judgers directly about ambiguity instead of collecting multiple drawn segmentations | ambiguity labeling · Weizmann | label agreement: matches labels for 79% of images with over 10x less effort (4.7 vs 50 seconds) | Improved |
| exclude ambiguous images from evaluation | segmentation algorithm benchmarking · Weizmann | top-performing method: 19 of the images are labeled as ambiguous; top method switches | Mixed |
| allocate redundant segmentations by predicted ambiguity instead of baselines | crowdsourcing pipeline · 800 images, region-based measure | effort to capture diversity: accelerates collection of ~51% of diversity by at least 27% | Improved |
| allocate redundant segmentations by predicted ambiguity instead of baselines | crowdsourcing pipeline · 800 images, boundary-based measure | effort to capture diversity: accelerates collection of ~53% of diversity by as much as 47% | Improved |
| capture most of total diversity (about 70%) | ambiguity-guided allocation · 800 images | gain over baselines: gains taper at ~70% of total diversity | No clear effect |
Closest papers by meaning
- Ambiguity Helps: Classification With Disagreements in Crowdsourced Annotations (CVPR 2016 · similarity 0.89)
- Does Progress On Object Recognition Benchmarks Improve Generalization on Crowdsourced, Global Data? (ICLR 2024 · similarity 0.86)
- FOCUS: Towards Universal Foreground Segmentation (AAAI 2025 · similarity 0.85)
- Flaws can be Applause: Unleashing Potential of Segmenting Ambiguous Objects in SAM (NeurIPS 2024 · similarity 0.85)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.84)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.84)
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.
