Research map / Detection and segmentation / Weakly supervised detection and segmentation
Improving Weakly Supervised Visual Grounding by Contrastive Knowledge Distillation
Liwei Wang, Jing Huang, Yin Li, Kun Xu, Zhengyuan Yang, Dong Yu
Check 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 region-phrase distillation to image-sentence NCE (ablation) | NCE grounding model · Flickr30K Entities | accuracy: 48.35 -> 50.96 | Improved |
| add region-phrase distillation to image-sentence NCE (ablation) | NCE grounding model · ReferItGame | accuracy: 26.63 -> 27.59 | Improved |
| add image-sentence NCE supervision to region-phrase distillation (ablation) | Distill grounding model · Flickr30K Entities | accuracy: 45.05 -> 50.96 | Improved |
| add image-sentence NCE supervision to region-phrase distillation (ablation) | Distill grounding model · ReferItGame | accuracy: 17.25 -> 27.59 | Improved |
| replace max-margin ranking loss with image-sentence NCE (ablation) | image-sentence matching model without distillation · Flickr30K Entities | accuracy: 42.11 -> 48.35 | Improved |
| replace max-margin ranking loss with image-sentence NCE (ablation) | image-sentence matching model without distillation · ReferItGame | accuracy: 22.94 -> 26.63 | Improved |
| add region-phrase distillation (ablation) | NCE model with VGG16 PV features and VGG16 CC detector · Flickr30K Entities | accuracy: 35.96 -> 40.38 (+4.42) | Improved |
| add region-phrase distillation (ablation) | NCE model with Res101 CC features and IRV2 OI detector · Flickr30K Entities | accuracy: 48.35 -> 50.96 (+2.61) | Improved |
| add region-phrase distillation (ablation) | NCE model with Res101 VG features and IRV2 OI detector · Flickr30K Entities | accuracy: 51.31 -> 53.10 (+1.79) | Improved |
| add region-phrase distillation (ablation) | NCE model with VGG16 PV features and VGG16 CC detector · ReferItGame | accuracy: 23.56 -> 24.52 (+0.96) | Improved |
| add region-phrase distillation (ablation) | NCE model with Res101 CC features and IRV2 OI detector · ReferItGame | accuracy: 26.63 -> 27.59 (+0.96) | Improved |
| add region-phrase distillation (ablation) | NCE model with Res101 VG features and IRV2 OI detector · ReferItGame | accuracy: 36.65 -> 38.39 (+1.74) | Improved |
Closest papers by meaning
- Contrastive Learning for Weakly Supervised Phrase Grounding (ECCV 2020 · similarity 0.92)
- Weakly-Supervised Visual Grounding of Phrases With Linguistic Structures (CVPR 2017 · similarity 0.92)
- Align2Ground: Weakly Supervised Phrase Grounding Guided by Image-Caption Alignment (ICCV 2019 · similarity 0.91)
- What is Where by Looking: Weakly-Supervised Open-World Phrase-Grounding without Text Inputs (NeurIPS 2022 · similarity 0.90)
- Knowledge Aided Consistency for Weakly Supervised Phrase Grounding (CVPR 2018 · similarity 0.90)
- Momentum Pseudo-Labeling for Weakly Supervised Phrase Grounding (AAAI 2025 · 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.
