Research map / Video / Video segmentation and tracking
Unsupervised Learning From Video to Detect Foreground Objects in Single Images
Ioana Croitoru, Simion-Vlad Bogolin, Marius Leordeanu
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 |
|---|---|---|---|
| replace VideoPCA teacher with trained student CNN | unsupervised foreground detector · ImageNet VID val | max F-measure: 41.83 -> 51.17 | Improved |
| add skip connections and HSV/derivative input channels (refined model) (ablation) | student CNN · ImageNet VID val | max F-measure: 51.17 -> 52.51 | Improved |
| add 10% soft-mask selection (ablation) | refined student CNN · ImageNet VID val | max F-measure: 52.51 -> 53.82 | Improved |
| keep only top 5% of masks instead of all (ablation) | refined student CNN · ImageNet VID val | max F-measure: 52.51 -> 53.20 | Improved |
| keep top 30% of masks (ablation) | refined student CNN · ImageNet VID val | max F-measure: 52.51 -> 53.67 | Improved |
| select 10% instead of 5% of masks (ablation) | refined student CNN · ImageNet VID val | max F-measure: 53.20 -> 53.82 | Improved |
| add unlabeled YTO videos to training (ablation) | baseline student CNN · ImageNet VID val | max F-measure: 51.17 -> 51.9 | Improved |
| add unlabeled YTO videos to training (ablation) | refined student, 10% selection · ImageNet VID val | max F-measure: 53.82 -> 54.53 | Improved |
| add YTO videos to training (ablation) | student CNN · Object Discovery CorLoc (3 classes avg) | CorLoc: 86.04 -> 85.83 | No clear effect |
| add YTO videos to training (ablation) | student CNN · Object Discovery segmentation | mean J: 61.25 -> 62.33 | Improved |
| add YTO videos to training (ablation) | student CNN · YouTube Objects v1 | CorLoc avg: 56.1 -> 61.6 | Improved |
| add skip connections (ablation) | student CNN · ImageNet VID val | performance: 1% | Improved |
Closest papers by meaning
- Unsupervised Learning of Foreground Object Segmentation (IJCV 2019 · similarity 0.97)
- Unsupervised Object Segmentation in Video by Efficient Selection of Highly Probable Positive Features (ICCV 2017 · similarity 0.92)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.91)
- Unsupervised Object Localization: Observing the Background To Discover Objects (CVPR 2023 · similarity 0.91)
- Learning Features by Watching Objects Move (CVPR 2017 · similarity 0.90)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.89)
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.
