AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Language-Aware Vision Transformer for Referring Segmentation

Zhao Yang, Jiaqi Wang, Xubing Ye, Yansong Tang, Kai Chen, Hengshuang Zhao, Philip H. S. Torr

TPAMI 2024 · Vision-language and multimodal › Cross-modal learning · approach: Contrastive vision-language alignment · 29 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 language pathway (ablation)LAVT · RefCOCO valoIoU: 70.78 -> 72.73Improved
add pixel-word attention module (spatial word features) instead of pooled sentence vector (ablation)LAVT · RefCOCO valoIoU: 71.03 -> 72.73Improved
add both language pathway and PWAM (ablation)LAVT · RefCOCO valoIoU: 68.82 -> 72.73Improved
use tanh instead of sigmoid in language gate (ablation)LAVT · RefCOCO valoIoU: 70.49 -> 72.73Improved
use InstanceNorm instead of no normalization in PWAM (ablation)LAVT · RefCOCO valoIoU: 70.66 -> 72.73Improved
use InstanceNorm instead of LayerNorm in PWAM (ablation)LAVT · RefCOCO valoIoU: 71.92 -> 72.73Improved
add language gate on F features (ablation)LAVT · RefCOCO valoIoU: 72.24 -> 72.73Improved
add language gate on E features (ablation)LAVT · RefCOCO valoIoU: 72.27 -> 72.06Hurt
add language gate on V features (ablation)LAVT · RefCOCO valoIoU: 72.29 -> 71.38Hurt
use PWAM instead of BCAM (ablation)LAVT · RefCOCO valoIoU: 70.19 -> 72.73Improved
use PWAM instead of GARAN attention (ablation)LAVT · RefCOCO valoIoU: 71.20 -> 72.73Improved
fuse by summation instead of concatenation (ablation)LAVT · RefCOCO valoIoU: 60.52 -> 72.24Improved

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