Research map / Detection and segmentation / Open-set and few-shot detection
Pull the Plug? Predicting If Computers or Humans Should Segment Images
Danna Gurari, Suyog Jain, Margrit Betke, Kristen Grauman
Check 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 |
|---|---|---|---|
| Predict segmentation quality from binary-mask features instead of CPMC features | Segmentation-quality prediction · BU-BIL | CC: 0.36 -> 0.61 | Improved |
| Predict segmentation quality from binary-mask features instead of CPMC features | Segmentation-quality prediction · Weizmann | MAE: 0.32 -> 0.24 | Improved |
| Use AlexNet CNN features for prediction instead of binary-mask features | Segmentation-quality prediction · BU-BIL, Weizmann, and IIS | CC: 0.61 -> -0.01; 0.64 -> -0.1; 0.68 -> -0.01 | Hurt |
| Train and test on the same dataset rather than cross-set training | Segmentation-quality prediction · BU-BIL | CC: 0.61 -> 0.69 | Improved |
| Train and test on the same dataset rather than cross-set training | Segmentation-quality prediction · IIS | CC: 0.68 -> 0.78 | Improved |
| Select computer initializations using predicted quality and allocate humans to the lowest-scoring images | PTP coarse-segmentation initialization · Weizmann, IIS, and BU-BIL; segmentation-tool initialization | human annotation effort eliminated: 44% | Improved |
| Use predicted-quality allocation rather than the Rectangle baseline | Lankton level sets initialization · Weizmann, IIS, and BU-BIL | segmentation quality: 20 percentage points | Improved |
| Use predicted-quality allocation rather than relying exclusively on human initialization | PTP coarse-segmentation initialization · Weizmann, IIS, and BU-BIL | segmentation quality: six percentage point improvement | Improved |
| Allocate humans to low-predicted-quality computer segmentations and retain computer results elsewhere | PTP fine-grained segmentation · Weizmann, IIS, and BU-BIL | human annotation time saved at comparable segmentation quality: 30-60 minutes less human annotation time | Improved |
| Use computer segmentations instead of crowd annotations when predictions favor the computer | PTP fine-grained segmentation · Last 100 images assigned human annotations | images where computer segmentation has higher Jaccard than crowd: 10% | Improved |
Closest papers by meaning
- Predicting How to Distribute Work Between Algorithms and Humans to Segment an Image Batch (IJCV 2019 · similarity 0.96)
- Clickstream Analysis for Crowd-Based Object Segmentation with Confidence (TPAMI 2017 · similarity 0.86)
- Supervised Evaluation of Image Segmentation and Object Proposal Techniques (TPAMI 2016 · similarity 0.86)
- Large-Scale Interactive Object Segmentation With Human Annotators (CVPR 2019 · similarity 0.86)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.84)
- TETRIS: Towards Exploring the Robustness of Interactive Segmentation (AAAI 2024 · 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.
