Research map / Detection and segmentation / Weakly supervised detection and segmentation
Cap2Det: Learning to Amplify Weak Caption Supervision for Object Detection
Keren Ye, Mingda Zhang, Adriana Kovashka, Wei Li, Danfeng Qin, Jesse Berent
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 text classifier label inference for unmatched captions (ablation) | ExactMatch · VOC 2007 (COCO captions) | mAP: 39.9 -> 43.1 | Improved |
| use text classifier instead of manual ExtendVocab synonyms (ablation) | EM + ExtendVocab · VOC 2007 (COCO captions) | mAP: 42.5 -> 43.1 | Improved |
| add GloVe pseudo labels (ablation) | ExactMatch · VOC 2007 (COCO captions) | mAP: 39.9 -> 40.5 | Improved |
| learn GloVe embeddings on COCO (ablation) | EM + GloVePseudo · VOC 2007 (COCO captions) | mAP: 40.5 -> 41.7 | Improved |
| add text classifier labels (ablation) | ExactMatch · VOC 2007 (Flickr30K captions) | mAP: 31.0 -> 33.6 | Improved |
| add manual ExtendVocab labels (ablation) | ExactMatch · VOC 2007 (Flickr30K captions) | mAP: 31.0 -> 29.3 | Hurt |
| add text classifier labels (ablation) | ExactMatch · COCO test-dev | AP@0.5: 19.7 -> 20.2 | Improved |
| add GloVe pseudo labels (ablation) | ExactMatch · COCO test-dev | AP@0.5: 19.7 -> 19.0 | Hurt |
| add OICR online refinement (1, 2, 3 times) (ablation) | basic detection network · VOC 2007 (image-level labels) | mAP: +24%, +29%, +30% | Improved |
| use Inception-V2 based architecture instead of VGG_M (ablation) | OICR with no refinement · VOC 2007 (image-level labels) | mAP: +27% | Improved |
| train text classifier on 100% instead of 5% of COCO (ablation) | text classifier · COCO val label precision | precision: 89% -> 92% | No clear effect |
| increase caption training data to 1M samples (ablation) | Cap2Det · VOC 2007 (Flickr data) | mAP: 54.4 mAP at 1 million samples (extrapolated) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- TS2C — TS2C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection (ECCV 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning to Overcome Noise in Weak Caption Supervision for Object Detection (TPAMI 2022 · similarity 0.92)
- W2N: Switching from Weak Supervision to Noisy Supervision for Object Detection (ECCV 2022 · similarity 0.90)
- Open-Vocabulary Object Detection Using Captions (CVPR 2021 · similarity 0.89)
- Weakly Supervised Open-Vocabulary Object Detection (AAAI 2024 · similarity 0.89)
- Weakly Supervised Few-Shot Object Detection with DETR (AAAI 2024 · similarity 0.88)
- Segment and Caption Anything (CVPR 2024 · similarity 0.87)
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.
