Research map / Detection and segmentation / Open-set and few-shot detection
Cut and Learn for Unsupervised Object Detection and Instance Segmentation
Read the paperCheck 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 |
|---|---|---|---|
| Use CutLER instead of the previous state-of-the-art method | Zero-shot object detection · 11-benchmark zero-shot average | AP50: 9.0 -> 24.3 | Helped |
| Use MaskCut to generate multiple initial masks (ablation) | CutLER · COCO | AP50 mask: 13.5 -> 15.8 | Helped |
| Use MaskCut to generate multiple initial masks (ablation) | CutLER · UVO | AP50 mask: 14.6 -> 19.3 | Helped |
| Add DropLoss to ignore low-overlap predicted regions (ablation) | CutLER with MaskCut · COCO | AP50 mask: 15.8 -> 16.6 | Helped |
| Add DropLoss to ignore low-overlap predicted regions (ablation) | CutLER with MaskCut · UVO | AP50 mask: 19.3 -> 20.9 | Helped |
| Add copy-paste augmentation (ablation) | CutLER with MaskCut and DropLoss · COCO | AP50 mask: 16.6 -> 17.7 | Helped |
| Add copy-paste augmentation (ablation) | CutLER with MaskCut and DropLoss · UVO | AP50 mask: 20.9 -> 21.5 | Helped |
| Add self-training (ablation) | CutLER with MaskCut, DropLoss, and copy-paste · COCO | AP50 mask: 17.7 -> 18.9 | Helped |
| Add self-training (ablation) | CutLER with MaskCut, DropLoss, and copy-paste · UVO | AP50 mask: 21.5 -> 22.8 | Helped |
| Use 4 rather than 3 self-training rounds (ablation) | CutLER · COCO | AP50 mask: 18.9 -> 18.9 | No clear effect |
| Use 4 rather than 3 masks per image (ablation) | MaskCut · COCO val2017 | AP50 mask: 17.7 -> 17.7 | No clear effect |
| Raise DropLoss IoU threshold from 0.01 to 0.1 (ablation) | CutLER · COCO val2017 | AP50 mask: 17.7 -> 14.4 | Hurt |
Closest papers by meaning
- Unsupervised Object Localization: Observing the Background To Discover Objects (CVPR 2023 · similarity 0.88)
- Unsupervised Learning of Edges (CVPR 2016 · similarity 0.88)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.86)
- CuVLER: Enhanced Unsupervised Object Discoveries through Exhaustive Self-Supervised Transformers (CVPR 2024 · similarity 0.85)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.85)
- 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.