AI Paper Atlas

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

Vision-Language Transformer and Query Generation for Referring Segmentation

Henghui Ding, Chang Liu, Suchen Wang, Xudong Jiang

ICCV 2021 · Vision-language and multimodal › Human and embodied modeling · approach: Referring expression comprehension · 303 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 transformer, QGM and QBM in place of seven conv layers (ablation)VLT · RefCOCO+ test BIoU: 44.28 -> 49.36Improved
add transformer, QGM and QBM in place of seven conv layers (ablation)VLT · RefCOCO+ test BPr@0.5: 49.54 -> 55.84Improved
generate queries with Query Generation Module instead of word features (ablation)VLT · RefCOCO+ test BIoU: 45.05 -> 49.36Improved
generate queries with Query Generation Module instead of fixed learned queries (ablation)VLT · RefCOCO+ test BIoU: 42.99 -> 49.36Improved
increase number of queries from 1 to 16 (ablation)VLT · RefCOCO+ test BIoU: 44.83 -> 49.36Improved
increase number of queries from 16 to 32 (ablation)VLT · RefCOCO+ test BIoU: 49.36 -> 49.27No clear effect
increase number of queries from 2 to 4 (ablation)VLT · RefCOCO+ test BIoU: 47.07 -> 46.79; Pr@0.5 52.85 -> 53.06Mixed
add Query Balance Module (ablation)VLT with 16 queries · RefCOCO+ test BIoU: 48.94 -> 49.36Improved
add Query Balance Module (ablation)VLT with 16 queries · RefCOCO+ test BPr@0.9: 12.99 -> 10.76Mixed
VLT vs best prior method CGANVLT · RefCOCO+ test BIoU: 44.06 -> 49.36Improved
VLT vs best prior method CGANVLT · RefCOCO+ test BIoU (RefCOCO val): 64.86 -> 65.65Improved
VLT vs best prior method CGANVLT · RefCOCO+ test BIoU (G-Ref test UMD): 51.69 -> 56.65Improved

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