Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Learning of Edges
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 edge detection using video-harvested motion edges instead of BSDS image-edge labels | SE · BSDS test | ODS: .746 -> .724 | Hurt |
| Train edge detection using video-harvested motion edges instead of BSDS image-edge labels | HE · BSDS test | ODS: .785 -> .748 | Hurt |
| Use video-harvested motion edges rather than the small VSB set of ground-truth motion edges | SE · BSDS test | ODS: .722 -> .724 | Helped |
| Use video-harvested motion edges rather than the small VSB set of ground-truth motion edges | HE · BSDS test | ODS: .745 -> .748 | Helped |
| Initialize HE with ImageNet pre-training rather than random initialization (ablation) | HE edge detector · BSDS test | ODS: .726 -> .748 | Helped |
| Remove deep supervision and use a single loss function (ablation) | HE · OIS (dataset not specified in the cited comparison) | ODS: .790 -> .785 | Hurt |
| Use video-trained SE edges instead of BSDS-trained SE edges for EpicFlow | EpicFlow · MPI-Sintel train | AEE: 3.686 -> 3.681 | No clear effect |
| Use video-trained HE edges instead of BSDS-trained HE edges for EpicFlow | EpicFlow · MPI-Sintel train | AEE: 3.608 -> 3.773 | Hurt |
| Pre-train VGG on video-trained edge detection rather than BSDS edge detection | Fast R-CNN with VGG · PASCAL VOC 2007 test | mAP: 42.1 -> 44.2 | Helped |
| Pre-train ZF on video-trained edge detection rather than BSDS edge detection | Fast R-CNN with ZF · PASCAL VOC 2007 test | mAP: 41.4 -> 41.1 | Hurt |
| Pre-train on edge detection rather than use no pre-training | Fast R-CNN with ZF · PASCAL VOC 2007 test | mAP: 38.2 -> 41.4 | Helped |
| Pre-train on ImageNet rather than video-trained edge detection | Fast R-CNN with ZF · PASCAL VOC 2007 test | mAP: 41.1 -> 58.6 | Helped |
Closest papers by meaning
- Learning Relaxed Deep Supervision for Better Edge Detection (CVPR 2016 · similarity 0.89)
- MEMO: Human-like Crisp Edge Detection Using Masked Edge Prediction (CVPR 2026 · similarity 0.89)
- Learning Features by Watching Objects Move (CVPR 2017 · similarity 0.88)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.87)
- SAUGE: Taming SAM for Uncertainty-Aligned Multi-Granularity Edge Detection (AAAI 2025 · similarity 0.87)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.85)
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.