Research map / Detection and segmentation / Open-set and few-shot detection
Learning to Find Common Objects Across Few Image Collections
Amirreza Shaban, Amir Rahimi, Shray Bansal, Stephen Gould, Byron Boots, Richard Hartley
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 |
|---|---|---|---|
| add learned unary potentials from the negative bag to pairwise-only inference (ablation) | greedy structured inference · COCO co-localization | success rate: 64.65 -> 65.34 | Improved |
| add learned unary potentials from the negative bag to pairwise-only inference (ablation) | greedy structured inference · ILSVRC2013 co-localization | success rate: 53.00 -> 55.18 | Improved |
| add pairwise potentials to unary-only selection (ablation) | greedy structured inference · COCO co-localization | success rate: 59.24 -> 65.34 | Improved |
| add pairwise potentials to unary-only selection (ablation) | greedy structured inference · ILSVRC2013 co-localization | success rate: 50.29 -> 55.18 | Improved |
| use learned softmax-weighted unary instead of no unary (ablation) | Ours · miniImageNet (N=8, B=5, 10 negatives) | success rate: 64.48 -> 72.49 | Improved |
| use learned temperature (SOFTMAX) instead of MEAN unary (ablation) | Ours · miniImageNet (N=8, B=5, 10 negatives) | success rate: 70.23 -> 72.49 | Improved |
| use learned temperature (SOFTMAX) instead of MAX unary (ablation) | Ours · miniImageNet (N=8, B=5, 10 negatives) | success rate: 71.76 -> 72.49 | Improved |
| use separate parameters for unary and pairwise relation modules instead of sharing (ablation) | Ours · miniImageNet (N=8, B=5, 10 negatives) | success rate: 69.35 -> 72.49 | Improved |
| replace cosine-similarity relation with learned relation network (ablation) | Ours · miniImageNet (N=8, B=5, 10 negatives) | success rate: 64.46 -> 72.49 | Improved |
| replace cosine-similarity relation with learned relation network (ablation) | Ours · miniImageNet (N=16, B=5, 10 negatives) | success rate: 66.78 -> 78.60 | Improved |
| use gated activation instead of ReLU in the relation embedding (ablation) | relation network · miniImageNet 5-way 1-shot | accuracy: 56.43 -> 57.80 | Improved |
| use greedy merge-and-prune inference instead of TRWS (ablation) | Ours · COCO co-localization | success rate: 65.04 -> 65.34 | No clear effect |
Closest papers by meaning
- Meta-Learning to Detect Rare Objects (ICCV 2019 · similarity 0.87)
- SMILe: Leveraging Submodular Mutual Information For Robust Few-Shot Object Detection (ECCV 2024 · similarity 0.85)
- SILCO: Show a Few Images, Localize the Common Object (ICCV 2019 · similarity 0.85)
- Multi-faceted Distillation of Base-Novel Commonality for Few-Shot Object Detection (ECCV 2022 · similarity 0.85)
- Unsupervised Image Matching and Object Discovery as Optimization (CVPR 2019 · similarity 0.85)
- Few-Shot Learning With Localization in Realistic Settings (CVPR 2019 · similarity 0.84)
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.
