Research map / Detection and segmentation / Weakly supervised detection and segmentation
StarNet: towards Weakly Supervised Few-Shot Object Detection
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 |
|---|---|---|---|
| Train the backbone end-to-end through the geometric StarNet head using image-level labels (ablation) | StarNet · ImageNetLOC-FS weakly supervised few-shot detection, 1-shot | AP (%) at IoU≥0.3: 2.1 -> 50.0 | Helped |
| Train the backbone end-to-end through the geometric StarNet head using image-level labels (ablation) | StarNet · CUB weakly supervised few-shot detection, 1-shot | AP (%) at IoU≥0.3: 3.5 -> 77.1 | Helped |
| Add geometric StarNet matching and localization | few-shot detection · ImageNetLOC-FS, 1-shot | AP (%) at IoU≥0.3: 32.4 -> 50.0 | Helped |
| Add geometric StarNet matching and localization | few-shot detection · CUB, 1-shot | AP (%) at IoU≥0.3: 60.7 -> 77.1 | Helped |
| Combine stage-1 geometric evidence with weighted stage-2 region features (ablation) | StarNet · CUB 1-shot, 5-way | accuracy (%): 76.74 -> 78.78 | Helped |
| Add back-projection-weighted pooling to stage-2 instead of unweighted pooling (ablation) | StarNet stage-2 · CUB 1-shot, 5-way | accuracy (%): 72.92 -> 76.74 | Helped |
| Add iterative NMS to expand matched regions (ablation) | Full StarNet · CUB 1-shot, 5-way | accuracy (%): 78.78 -> 79.58 | Helped |
| Use separate backbones for the two StarNet stages rather than a shared backbone (ablation) | StarNet · ImageNetLOC-FS 1-shot | accuracy (%): 61.0 -> 63.0 | Helped |
| Use separate backbones for the two StarNet stages rather than a shared backbone (ablation) | StarNet · CIFAR-FS 1-shot | accuracy (%): 73.65 -> 73.34 | Hurt |
| Apply StarNet to few-shot classification instead of relying on standard feature matching | few-shot classification · CUB 1-shot | accuracy (%): 75.01 -> 79.58 | Helped |
| Apply StarNet to few-shot classification instead of standard feature matching | few-shot classification · miniImageNet 1-shot | accuracy (%): 63.85 -> 63.4 | No clear effect |
Closest papers by meaning
- Label, Verify, Correct: A Simple Few Shot Object Detection Method (CVPR 2022 · similarity 0.89)
- Meta Faster R-CNN: Towards Accurate Few-Shot Object Detection with Attentive Feature Alignment (AAAI 2022 · similarity 0.89)
- Sylph: A Hypernetwork Framework for Incremental Few-Shot Object Detection (CVPR 2022 · similarity 0.88)
- Few-Shot Learning With Localization in Realistic Settings (CVPR 2019 · similarity 0.88)
- Revisiting Few-Shot Object Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · 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.