AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Iterative Shrinking for Referring Expression Grounding Using Deep Reinforcement Learning

Mingjie Sun, Jimin Xiao, Eng Gee Lim

CVPR 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Object-level representation learning · 35 citations (OpenAlex)

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 changedWhereMeasuredEffect
use iterative shrinking with an RL agent instead of feature-point matchingproposal-free referring expression grounding · RefCOCOg testaccuracy: 65.73 -> 70.05Improved
use iterative shrinking with an RL agent instead of feature-point matchingproposal-free referring expression grounding · RefCOCO testA and testBaccuracy: 78.93 -> 74.27; 73.09 -> 68.10Mixed
use iterative shrinking with an RL agentreferring expression grounding · RefCOCOg, RefCOCO+, RefCOCO averageaccuracy: 1.33%Improved
add triad-based linguistic features instead of BERT features (ablation)ISREG · RefCOCO testAaccuracy: 67.52 -> 74.27Improved
add triad-based linguistic features instead of BERT features (ablation)ISREG · RefCOCOg testaccuracy: 60.17 -> 70.05Improved
add multi-scale visual features (ablation)ISREG · RefCOCO testAaccuracy: 73.33 -> 74.27Improved
add multi-scale visual features (ablation)ISREG · RefCOCOg testaccuracy: 66.35 -> 70.05Improved
add spatial features to the state (ablation)ISREG · RefCOCO testAaccuracy: 62.42 -> 74.27Improved
add spatial features to the state (ablation)ISREG · RefCOCOg testaccuracy: 54.88 -> 70.05Improved
use adaptive rather than fixed shrinking stride (ablation)ISREG · RefCOCO testA and RefCOCOg testaccuracy: 45.70 -> 74.27; 39.46 -> 70.05Improved
add bounding-box refinement (ablation)ISREG · RefCOCO testA and RefCOCOg testaccuracy: 70.18 -> 74.27; 65.62 -> 70.05Improved
use reinforcement learning rather than supervised training (ablation)ISREG · RefCOCO testA and RefCOCOg testaccuracy: 63.84 -> 74.27; 61.36 -> 70.05Improved

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