Research map / Detection and segmentation / Open-set and few-shot detection
Clickstream Analysis for Crowd-Based Object Segmentation with Confidence
Eric Heim, Alexander Seitel, Jonas Andrulis, Fabian Isensee, Christian Stock, Tobias Roß, Lena Maier-Hein
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 |
|---|---|---|---|
| filter crowd segmentations with clickstream-based quality estimation (eps_t=0.9, lambda=3) | Crowd segmentation pool · PASCAL VOC cat+car | mean DSC: 0.80 -> 0.93 | Improved |
| filter crowd segmentations with clickstream-based quality estimation | Crowd segmentation pool · PASCAL VOC cat+car | median DSC: 0.91 -> 0.95 | Improved |
| replace conventional majority voting with confidence-weighted majority voting | Majority voting · PASCAL VOC car/cat | annotations needed for median DSC 0.95: reduced by 75% | Improved |
| add segmentation quality estimation to STAPLE | Native STAPLE · PASCAL VOC car/cat | annotations needed for median DSC 0.95: reduced by 73% | Improved |
| add clickstream features to image-based features (ablation) | Quality estimator using image-only features · PASCAL VOC cat and car | mean estimation error: error approximately twice as high without clickstream | Improved |
| reduce feature set to 6 features with CIFE instead of all 53 (ablation) | Random forest regressor · PASCAL VOC cat | mean estimation error: 0.06 -> 0.06 | No clear effect |
| reduce feature set to 6 features with ICAP instead of all 53 (ablation) | Random forest regressor · PASCAL VOC car | mean estimation error: 0.10 -> 0.10 | No clear effect |
| use SFS-selected features instead of all 53 (ablation) | Random forest regressor · PASCAL VOC car | mean estimation error: 0.10 -> 0.11 | Hurt |
| use cross-category training (animal to vehicle or vice versa) | Quality estimator · MTurk COCO classes | mean estimation error: below 0.1; annotations reduced by 50% | Mixed |
| train on class farther from target class (ablation) | Quality estimator · MTurk COCO classes | estimation error | Hurt |
| apply quality-weighted merging for lambda>=2 instead of manual grading | Manual grading cost approach (COCO) · Cost estimate, ~25% spam | annotation cost | Improved |
| use proposed method instead of majority voting with quality control tasks | Majority voting baseline · Cost estimate, 30% spam | annotation cost | Improved |
Closest papers by meaning
- Lean Crowdsourcing: Combining Humans and Machines in an Online System (CVPR 2017 · similarity 0.88)
- Lean Multiclass Crowdsourcing (CVPR 2018 · similarity 0.87)
- Learning From Noisy Singly-labeled Data (ICLR 2018 · similarity 0.87)
- Full or Weak Annotations? An Adaptive Strategy for Budget-Constrained Annotation Campaigns (CVPR 2023 · similarity 0.86)
- Regressor-Segmenter Mutual Prompt Learning for Crowd Counting (CVPR 2024 · similarity 0.84)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · 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.
