Research map / Detection and segmentation / Weakly supervised detection and segmentation
Visual and Semantic Knowledge Transfer for Large Scale Semi-Supervised Object Detection
Yuxing Tang, Josiah Wang, Xiaofang Wang, Boyang Gao, Emmanuel Dellandréa, Robert Gaizauskas, Liming Chen
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 |
|---|---|---|---|
| replace LSDA nearest-neighbor adaptation with visual-similarity weighted transfer (weighted-100) | LSDA AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 16.33 -> 19.02 | Improved |
| use semantic relatedness with sparse reconstruction instead of LSDA | LSDA AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 16.33 -> 19.04 | Improved |
| combine visual and semantic transfer (mixture) (ablation) | visual transfer AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 19.02 -> 20.03 | Improved |
| add transferred bounding-box regression (ablation) | mixture transfer AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 20.03 -> 21.88 | Improved |
| weight neighbors by similarity instead of simple averaging (ablation) | visual transfer AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 17.38 -> 19.02 | Improved |
| use AutoExtend synset embeddings instead of naive word2vec embeddings (ablation) | semantic transfer weighted-100 · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 17.83 -> 18.32 | Improved |
| use AutoExtend synset embeddings instead of WordNet path similarity (ablation) | semantic transfer weighted-100 · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 17.08 -> 18.32 | Improved |
| use AutoExtend synset embeddings instead of Lin similarity (ablation) | semantic transfer weighted-100 · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 17.31 -> 18.32 | Improved |
| use sparse reconstruction (<=20) instead of weighted-100 for semantic transfer (ablation) | semantic transfer AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 18.32 -> 19.04 | Improved |
| increase visual-transfer neighbors from 5 to 10 (weighted) (ablation) | visual transfer AlexNet · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 17.59 -> 18.41 | Improved |
| swap AlexNet for ResNet-50 backbone | mixture transfer · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 20.03 -> 26.77 | Improved |
| use end-to-end Fast R-CNN BB regression instead of 2-stage (ablation) | Fast R-CNN ResNet-50 · ILSVRC2013 val2 (weakly labeled 100 categories) | mAP: 29.09 -> 29.71 | Improved |
Closest papers by meaning
- Large Scale Semi-Supervised Object Detection Using Visual and Semantic Knowledge Transfer (CVPR 2016 · similarity 0.99)
- Revisiting Knowledge Transfer for Training Object Class Detectors (CVPR 2018 · similarity 0.91)
- CaT: Weakly Supervised Object Detection With Category Transfer (ICCV 2021 · similarity 0.91)
- Mixed Supervised Object Detection by Transferring Mask Prior and Semantic Similarity (NeurIPS 2021 · similarity 0.90)
- Universal-RCNN: Universal Object Detector via Transferable Graph R-CNN (AAAI 2020 · similarity 0.89)
- Detecting 11K Classes: Large Scale Object Detection Without Fine-Grained Bounding Boxes (ICCV 2019 · 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.
