Research map / Detection and segmentation / CNN features
OS2D: One-Stage One-Shot Object Detection by Matching Anchor Features
Anton Osokin, Denis Sumin, Vasily Lomakin
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 |
|---|---|---|---|
| use RLL loss instead of contrastive loss (ablation) | OS2D V1 · GroZi-3.2k val-new-cl | mAP: 81.8 -> 85.7 | Improved |
| add recognition target remapping (ablation) | OS2D V1 · GroZi-3.2k val-new-cl | mAP: 85.7 -> 87.0 | Improved |
| add hard patch mining (ablation) | OS2D V1 · GroZi-3.2k val-new-cl | mAP: 87.0 -> 86.5 | No clear effect |
| add hard patch mining (ablation) | OS2D V2 · GroZi-3.2k val-new-cl | mAP: 89.5 -> 90.6 | Improved |
| initialize TransformNet from Rocco et al. weights (init transform) (ablation) | OS2D V2 RLL + remap · GroZi-3.2k val-new-cl | mAP: 87.0 -> 88.0 | Improved |
| zero out the localization loss (ablation) | OS2D V2 · GroZi-3.2k val-new-cl | mAP: 88.0 -> 89.5 | Improved |
| use V2 (full affine, weak supervision) instead of V1 TransformNet | OS2D · GroZi-3.2k val-new-cl | mAP: 88.7 -> 90.6 | Improved |
| pretrain feature extractor on ImageNet classification instead of random init (ablation) | OS2D V2-train · GroZi-3.2k val-new-cl | mAP: 67.9 -> 89.0 | Improved |
| initialize from COCO detection instead of ImageNet classification (ablation) | OS2D V2-train · GroZi-3.2k val-new-cl | mAP: 89.5 -> 88.1 | Hurt |
| replace batch norm with group norm in the converted ImageNet model (ablation) | OS2D V2-train · GroZi-3.2k val-new-cl | mAP: 89.5 -> 87.1 | Hurt |
| add learned transformation model (vs sliding window with target aspect ratio) (ablation) | OS2D · GroZi-3.2k paste-v | mAP: 65.1 -> 73.3 | Improved |
| use one-shot matching OS2D instead of detector plus retrieval | OS2D V2-train · GroZi-3.2k paste-v | mAP: 44.3 -> 73.3 | Improved |
Closest papers by meaning
- Learning Rich Features at High-Speed for Single-Shot Object Detection (ICCV 2019 · similarity 0.88)
- TOOD: Task-Aligned One-Stage Object Detection (ICCV 2021 · similarity 0.88)
- One Shot Detection with Laplacian Object and Fast Matrix Cosine Similarity (TPAMI 2016 · similarity 0.87)
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.87)
- OSOP: A Multi-Stage One Shot Object Pose Estimation Framework (CVPR 2022 · similarity 0.87)
- Cut and Learn for Unsupervised Object Detection and Instance Segmentation (CVPR 2023 · similarity 0.86)
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.
