Research map / Detection and segmentation / Open-set and few-shot detection
Predicting How to Distribute Work Between Algorithms and Humans to Segment an Image Batch
Danna Gurari, Yinan Zhao, Suyog Dutt Jain, 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 |
|---|---|---|---|
| use the proposed quality predictor instead of CPMC | quality predictor · BU-BIL (cross-dataset) | Pearson CC: 0.18 -> 0.63 | Improved |
| train on a single dataset instead of cross-dataset | quality predictor · all four datasets | Pearson CC: ~0.60 -> ~0.85 | Improved |
| add mask-based features to intensity-only features (ablation) | quality predictor · BU-BIL (cross-dataset) | Pearson CC: -0.10 -> 0.63 | Improved |
| add intensity features to mask-only features (ablation) | quality predictor · Weizmann (cross-dataset) | Pearson CC: 0.65 -> 0.61 | Hurt |
| add intensity features to mask-only features (ablation) | quality predictor · IIS (single-dataset) | Pearson CC: 0.73 -> 0.83 | Improved |
| add intensity features to mask-only features (ablation) | quality predictor · BU-BIL (single-dataset) | Pearson CC: 0.88 -> 0.87 | Hurt |
| predict best of 14 algorithm segmentations instead of Rectangle initialization | PTP initialization · All images, Lankton level set, 0% human | segmentation quality: 31 percentage points | Improved |
| predict best of 14 algorithm segmentations instead of Rectangle initialization | PTP initialization · All images, GrabCut, 0% human | segmentation quality: 15 percentage points | Improved |
| use non-linear regression trees over 14 options instead of Linear model over 8 options | PTP initialization · Weizmann, Lankton level set | segmentation quality: 14 percentage points | Improved |
| use predicted initialization instead of Rectangle with GrabCut | PTP initialization · IIS, GrabCut | segmentation quality: approximately two percentage points | Hurt |
| add refinement algorithm to coarse segmentation (ablation) | PTP initialization · All images, Lankton level set | segmentation quality: over 20 percentage points | Improved |
| add GrabCut refinement to coarse segmentation (ablation) | PTP initialization · All images, GrabCut | segmentation quality: 26 percentage improvement | Improved |
Closest papers by meaning
- Pull the Plug? Predicting If Computers or Humans Should Segment Images (CVPR 2016 · similarity 0.96)
- Supervised Evaluation of Image Segmentation and Object Proposal Techniques (TPAMI 2016 · similarity 0.87)
- Large-Scale Interactive Object Segmentation With Human Annotators (CVPR 2019 · similarity 0.86)
- FOCUS: Towards Universal Foreground Segmentation (AAAI 2025 · similarity 0.84)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.84)
- A Generative Approach at the Instance-Level for Image Segmentation Under Limited Training Data Conditions (Student Abstract) (AAAI 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.
