AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Contrastive Learning for Weakly Supervised Phrase Grounding

Tanmay Gupta, Arash Vahdat, Gal Chechik, Xiaodong Yang, Jan Kautz, Derek Hoiem

ECCV 2020 · Vision-language and multimodal › Human and embodied modeling · approach: Visual grounding transformers · 127 citations (OpenAlex)

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 changedWhereMeasuredEffect
add pretrained BERT word representations instead of random-init BERT (ablation)Ours, no negatives · Flickr30K Entities testpointing accuracy: 57.37 -> 66.89Improved
add pretrained BERT word representations instead of random-init BERT (ablation)Ours, no negatives · Flickr30K Entities testR@1: 25.66 -> 35.74Improved
add random negative captions from training data (ablation)Ours, pretrained BERT · Flickr30K Entities testpointing accuracy: 66.89 -> 66.92No clear effect
add random negative captions from training data (ablation)Ours, pretrained BERT · Flickr30K Entities testR@1: 35.74 -> 36.32No clear effect
use contextually plausible negatives instead of random negatives (ablation)Ours, pretrained BERT · Flickr30K Entities testpointing accuracy: 66.92 -> 74.91Improved
use contextually plausible negatives instead of random negatives (ablation)Ours, pretrained BERT · Flickr30K Entities testR@1: 36.32 -> 48.05Improved
exclude near-synonyms and hypernyms from plausible negatives (ablation)Ours, pretrained BERT · Flickr30K Entities testR@1: 48.05 -> 51.67Improved
exclude near-synonyms and hypernyms from plausible negatives (ablation)Ours, pretrained BERT · Flickr30K Entities testpointing accuracy: 74.91 -> 76.74Improved
train on Flickr30K Entities instead of COCOOurs · Flickr30K Entities testpointing accuracy: 76.74 -> 74.94Hurt
use Ours instead of Align2Ground with same visual featuresOurs (COCO train) · Flickr30K Entities testpointing accuracy: 71.00 -> 76.74Improved
use one or two layer MLP key-value modules instead of linear (ablation)Ours · Flickr30K Entities testgrounding performanceImproved
maximize InfoNCE bound to convergence rather than stopping early (ablation)Ours · Flickr30K Entities testphrase grounding performanceHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy