AI Paper Atlas

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

Modeling Context in Referring Expressions

Licheng Yu, Patrick Poirson, Shan Yang, Alexander C. Berg, Tamara L. Berg

ECCV 2016 · Vision-language and multimodal › Human and embodied modeling · approach: Multimodal retrieval-augmented modeling · 1,205 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 global image context (ablation)Baseline · RefCOCO testBcomprehension accuracy: 66.31% -> 64.21%Hurt
use cropped window around target (scale 4) instead of global context (ablation)Baseline · RefCOCO testBcomprehension accuracy: 64.21% -> 68.56%Improved
add visual comparison features (visdif) (ablation)Baseline · RefCOCO testBcomprehension accuracy: 64.21% -> 71.19%Improved
add visual comparison features (visdif) (ablation)MMI · RefCOCO+ testBcomprehension accuracy: 51.23% -> 55.62%Improved
add visual comparison features (visdif) (ablation)Baseline · RefCOCOg valcomprehension accuracy: 55.16% -> 59.25%Improved
add visual comparison features (visdif) (ablation)Baseline(det) · RefCOCO testB (detections)comprehension accuracy: 48.48% -> 50.80%Improved
compare only to same-category objects instead of different or all objects (ablation)visdif · RefCOCO/RefCOCO+comprehension accuracyImproved
add MMI training (ablation)Baseline · RefCOCO testAcomprehension accuracy: 63.15% -> 71.72%Improved
add visual comparison (visdif) (ablation)Baseline · RefCOCO testABleu 1: 0.477 -> 0.505Improved
add MMI training (ablation)Baseline · RefCOCO+ testABleu 1: 0.391 -> 0.370Hurt
add MMI training (ablation)Baseline · RefCOCO testAhuman-judged accuracy: 62.42% -> 65.76%Improved
add LSTM tying across objects (ablation)Baseline · RefCOCO testBhuman-judged accuracy: 64.99% -> 68.34%Improved

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