Research map / Detection and segmentation / Weakly supervised detection and segmentation
UniT: Unified Knowledge Transfer for Any-Shot Object Detection and Segmentation
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 |
|---|---|---|---|
| Transfer weak-to-strong refinements using learned lingual and visual similarities, including base-class regressors | UniT · MS-COCO 2015 novel classes (k=0) | AP50: 14.4 -> 16.7 | Helped |
| Use UniT transfer with novel-class instance examples | MS-COCO 2017 novel classes · 12 shots | mAP: 14.1 -> 14.7 | Helped |
| Add similarity-weighted base classifier and regressor transfer (ablation) | Weakly supervised detector on VOC split 1 · VOC novel classes, k=0 | AP50: 63.9 -> 75.7 | Helped |
| Replace similarity-weighted transfer with simple averaging of base refinements (ablation) | Weakly supervised detector on VOC split 1 · VOC novel classes, k=0 | AP50: 63.9 -> 63.7 | No clear effect |
| Add lingual similarity for classifier transfer (ablation) | Weakly supervised detector on VOC split 1 · VOC novel classes, k=0 | AP50: 63.9 -> 64.1 | Helped |
| Add visual similarity to lingual classifier transfer (ablation) | Weakly supervised detector on VOC split 1 · VOC novel classes, k=0 | AP50: 64.1 -> 64.9 | Helped |
| Add visual and lingual similarity weighting (ablation) | UniT on MS-COCO novel classes · k=0, box detection | AP50: +1.4 AP50 | Helped |
| Add transfer from base-class regressors (ablation) | UniT on MS-COCO novel classes · k=0, box detection | AP50: +7 AP50 | Helped |
| Add transfer from base-class segmentation heads (ablation) | UniT on MS-COCO novel classes · k=0, instance segmentation | mask AP50: +7.5 | Helped |
| Use only weak image-level annotations rather than mixing them with instance annotations at the same budget (ablation) | UniT budget setting on VOC split 1 · 10-instance-equivalent annotation budget | AP50: 52.8 ± 1.0 -> 59.0 ± 1.5 | Helped |
Closest papers by meaning
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.90)
- Learning Instance Activation Maps for Weakly Supervised Instance Segmentation (CVPR 2019 · similarity 0.90)
- Universal Weakly Supervised Segmentation by Pixel-to-Segment Contrastive Learning (ICLR 2021 · similarity 0.89)
- Learning to Segment Every Thing (CVPR 2018 · similarity 0.89)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.89)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.89)
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.