Research map / Detection and segmentation / Open-set and few-shot detection
Learning Features by Watching Objects Move
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train a ConvNet to predict single-frame object masks from unsupervised motion-segmentation labels | AlexNet representation · PASCAL VOC 2012 object detection, including 150-image training | object detection AP (%): 48.6 vs 49.9; 10.2 vs 6.7 | Helped |
| Train on YFCC video frames instead of ImageNet images (ablation) | Context representation · PASCAL VOC 2012 object detection, >c5 | object detection AP (%): 33.4% mean AP ... compared to 33.2% | No clear effect |
| Use ConvNet mask predictions instead of raw uNLC segmentations | Unsupervised motion segmentation · DAVIS, FBMS and VSB | Mean IoU (%): 13.1 -> 24.8 | Helped |
| Add boundary noise to training masks (ablation) | COCO-supervised segmentation representation · PASCAL VOC 2007 object detection | object detection AP | No clear effect |
| Truncate training masks, including truncation up to 50% (ablation) | COCO-supervised segmentation representation · PASCAL VOC 2007 object detection | object detection AP: up to 50% | No clear effect |
| Reduce the amount of training data for mask prediction (ablation) | COCO-supervised segmentation representation · PASCAL VOC 2007 object detection | object detection AP | Hurt |
| Increase the number of video frames used for representation training (ablation) | Motion Masks representation · PASCAL VOC 2007 object detection, conv layers frozen | mean AP: 396K frames; 1.6M frames; 27M frames | Helped |
| Fine-tune the entire network rather than freezing some layers (ablation) | Object detection representation transfer · PASCAL VOC 2012, 150 training images | object detection AP: 150 | Hurt |
Closest papers by meaning
- Unsupervised Multi-Object Segmentation by Predicting Probable Motion Patterns (NeurIPS 2022 · similarity 0.91)
- Bootstrapping Objectness From Videos by Relaxed Common Fate and Visual Grouping (CVPR 2023 · similarity 0.90)
- The Emergence of Objectness: Learning Zero-shot Segmentation from Videos (NeurIPS 2021 · similarity 0.90)
- Learning Motion Patterns in Videos (CVPR 2017 · similarity 0.90)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · similarity 0.89)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.88)
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.