AI Paper Atlas

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

Visual Semantic Reasoning for Image-Text Matching

Kunpeng Li, Yulun Zhang, Kai Li, Yuanyuan Li, Yun Fu

ICCV 2019 · Vision-language and multimodal › Human and embodied modeling · approach: Language-grounded scene text and structure · 644 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 region relationship reasoning (GCN) layer before mean-pooling (ablation)Mean-pool · MS-COCO 1K testCaption retrieval R@1: 64.3 -> 68.5Improved
add region relationship reasoning (GCN) layer before mean-pooling (ablation)Mean-pool · MS-COCO 1K testImage retrieval R@1: 49.2 -> 56.8Improved
replace mean-pooling with global semantic reasoning (GRU) (ablation)Mean-pool · MS-COCO 1K testCaption retrieval R@1: 64.3 -> 72.3Improved
replace mean-pooling with global semantic reasoning (GRU) (ablation)Mean-pool · MS-COCO 1K testImage retrieval R@1: 49.2 -> 59.6Improved
add one RRR layer before GSR (ablation)GSR · MS-COCO 1K testCaption retrieval R@1: 72.3 -> 75.3Improved
increase RRR layers from 1 to 4 (ablation)1RRR + GSR · MS-COCO 1K testCaption retrieval R@1: 75.3 -> 76.2Improved
increase RRR layers from 1 to 4 (ablation)1RRR + GSR · MS-COCO 1K testImage retrieval R@1: 62.1 -> 62.8Improved
add the text generation loss to joint training (ablation)4RRR + GSR · MS-COCO 1K testCaption retrieval R@1: 74.6 -> 76.2Improved
add the text generation loss to joint training (ablation)4RRR + GSR · MS-COCO 1K testImage retrieval R@1: 61.2 -> 62.8Improved
order regions by detection confidence instead of randomly (ablation)VSRN · MS-COCO 1K testCaption retrieval R@1: 75.1 -> 76.2Improved
order regions by confidence instead of bounding box size (ablation)VSRN · MS-COCO 1K testCaption retrieval R@1: 75.8 -> 76.2No clear effect
use VSRN instead of SCANSCAN · MS-COCO 1K testCaption retrieval R@1: 72.7 -> 76.2Improved

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