Research map / Detection and segmentation / Weakly supervised detection and segmentation
DOCK: Detecting Objects by transferring Common-sense Knowledge
Krishna Kumar Singh, Santosh Divvala, Ali Farhadi, Yong Jae Lee
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 attribute common-sense (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 12.2 | Improved |
| add spatial common-sense (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 13.0 | Improved |
| add similarity common-sense (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 13.7 | Improved |
| combine attribute, spatial and similarity common-sense (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 14.1 | Improved |
| transfer bounding box regressors from similar source classes (ablation) | joint model · MS COCO val (60 target classes) | mAP: 14.1 -> 14.4 | Improved |
| use common-sense only as test-time post-processing (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 11.8 (14.1 with Y in training) | Mixed |
| transfer common-sense during training instead of test-time post-processing (ablation) | DOCK · MS COCO val (60 target classes) | mAP: 11.8 -> 14.1 | Improved |
| concatenate source detection and attribute probabilities to fc7 features (ablation) | base detection network · MS COCO val (60 target classes) | mAP: 11.7 -> 12.7 | Improved |
| use word2vec instead of visual similarity for similarity common-sense (ablation) | similarity common-sense · MS COCO val (60 target classes) | mAP: 12.1 -> 13.7 | Improved |
| use word2vec instead of WordNet treepath similarity (ablation) | similarity common-sense · MS COCO val (60 target classes) | mAP: 13.1 -> 13.7 | Improved |
| replace LSDA transfer with DOCK common-sense transfer | LSDA · MS COCO val (60 target classes) | mAP: 4.6 -> 14.4 | Improved |
| initialize from source-class Fast-RCNN weights and fine-tune with image labels instead of DOCK | fine-tuned detection network · MS COCO val (60 target classes) | mAP: 14.4 -> 10.8 | Hurt |
Closest papers by meaning
- Visual and Semantic Knowledge Transfer for Large Scale Semi-Supervised Object Detection (TPAMI 2017 · similarity 0.87)
- Revisiting Knowledge Transfer for Training Object Class Detectors (CVPR 2018 · similarity 0.86)
- Cyclic Contrastive Knowledge Transfer for Open-Vocabulary Object Detection (ICLR 2025 · similarity 0.85)
- Rex-Thinker: Grounded Object Referring via Chain-of-Thought Reasoning (ICLR 2026 · similarity 0.85)
- DoGA: Enhancing Grounded Object Detection via Grouped Pre-Training with Attributes (AAAI 2025 · similarity 0.85)
- Seeing the Unseen: Visual Common Sense for Semantic Placement (CVPR 2024 · similarity 0.85)
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.
