AI Paper Atlas

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

SegVG: Transferring Object Bounding Box to Segmentation for Visual Grounding

Weitai Kang, Gaowen Liu, Mubarak Shah, Yan Yan

ECCV 2024 · Detection and segmentation › Open-set and few-shot detection · approach: Amodal and open-world instance segmentation · 17 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 Triple Alignment to baseline (ablation)Backbones + Encoder + Decoder · RefCOCOg-umd testaccuracy (%): 66.97 -> 75.75Improved
add multi-layer multi-task segmentation supervision (N=1) (ablation)Backbones + Encoder + Decoder · RefCOCOg-umd testaccuracy (%): 66.97 -> 72.61Improved
increase segmentation queries from 1 to 5 (ablation)MMDecoder · RefCOCOg-umd testaccuracy (%): 72.61 -> 76.21Improved
increase segmentation queries beyond 5 (7, 9) (ablation)MMDecoder · RefCOCOg-umd testaccuracy (%): 76.21 -> 74.38 (N=7), 73.91 (N=9)Hurt
combine Triple Alignment with multi-task decoder (ablation)SegVG · RefCOCOg-umd testaccuracy (%): 76.21 -> 77.29Improved
keep multimodal encoder after Triple Alignment (ablation)SegVG · RefCOCOg-umd testaccuracy (%): 76.65 -> 77.29Improved
include queries in Triple Alignment instead of text-vision bidirectional alignment only (ablation)SegVG · RefCOCOg-umd testaccuracy (%): 76.37 -> 77.29Improved
add segmentation-derived confidence as focal factor on other losses (ablation)SegVG · RefCOCOg-umd testaccuracy (%): 76.83 -> 77.29Improved
use bbox2seg masks instead of pseudo segmentation labels from a RES model (ablation)SegVG · ReferItGame testaccuracy (%): 73.76 -> 75.06Improved
choose which segmentation query yields the confidence score (ablation)SegVG · RefCOCOg-umd testAP50: 84.57 to 84.78 across queriesNo clear effect
raise confidence threshold at inferenceSegVG · RefCOCOg-umd testaccuracy and mean IoUImproved

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