AI Paper Atlas

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

Self-Supervised Relationship Probing

Jiuxiang Gu, Jason Kuen, Shafiq Joty, Jianfei Cai, Vlad I. Morariu, Handong Zhao, Tong Sun

NeurIPS 2020 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 8 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 data augmentation to pretraining corpus (ablation)SSRPShare · NLVR2 devaccuracy: 60.53 -> 61.67Improved
add data augmentation to pretraining corpus (ablation)SSRPVisual · NLVR2 devaccuracy: 69.92 -> 70.75Improved
add data augmentation to pretraining corpus (ablation)SSRPCross · NLVR2 devaccuracy: 74.35 -> 74.48Improved
use cross-attention (SSRPCross) instead of visual-only attention (ablation)SSRPVisual · NLVR2 devaccuracy: 72.03 -> 75.71Improved
use separate visual attention (SSRPVisual) instead of shared (ablation)SSRPShare · NLVR2 devaccuracy: 64.25 -> 72.03Improved
add language relationships Rw to falign (ablation)SSRPCross · NLVR2 devaccuracy: 74.48 -> 74.68Improved
add visual relationships Rv to falign (ablation)SSRPCross · NLVR2 devaccuracy: 74.48 -> 74.25No clear effect
combine visual and textual relationships Rv+Rw (ablation)SSRPCross · NLVR2 devaccuracy: 74.48 -> 75.71Improved
combine visual and textual relationships Rv+Rw (ablation)SSRPShare · NLVR2 devaccuracy: 61.67 -> 64.25Improved
combine visual and textual relationships Rv+Rw (ablation)SSRPVisual · NLVR2 devaccuracy: 70.75 -> 72.03Improved
compare SSRPCross against prior pretraining modelsSSRPCross · NLVR2 devaccuracy: 75.71 vs 74.9 (LXMERT)Improved
use SSRPVisual refined object features in Up-Down captionerUp-Down · MSCOCO test splitCIDEr: 120.8 -> 126.7Improved

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