AI Paper Atlas

Research map / Vision-language and multimodal / Visual question answering

Contrast and Classify: Training Robust VQA Models

Yash Kant, Abhinav Moudgil, Dhruv Batra, Devi Parikh, Harsh Agrawal

ICCV 2021 · Vision-language and multimodal › Visual question answering · approach: Universal VQA models · 34 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 ConClaT alternate contrastive+CE training with BT dataMMT baseline · VQA-RephrasingsConsensus Score CS(4) (%): +1.63%Improved
add ConClaT alternate contrastive+CE training with BT dataMMT baseline · VQA 2.0 valVQA accuracy (%): 66.31 -> 66.98Improved
add ConClaT with VQG-generated paraphrasesMMT + VQG data · VQA-RephrasingsConsensus Score CS(4) (%): +0.46%Improved
add back-translated paraphrase data with CE onlyMMT · VQA-RephrasingsConsensus Score CS(4) (%): +0.54%Improved
train jointly instead of alternately (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): -0.36%Hurt
use pretrain-finetune instead of alternate training (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): -4.79%Hurt
replace L_SSCL with dynamic-margin triplet loss (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): 53.99 -> 53.10Hurt
add scaling factor to supervised contrastive loss (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): 53.36 -> 53.99Improved
use dynamic scaling instead of constant scaling (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): 53.99 vs 53.92No clear effect
use curated question+image+random negatives instead of random negatives (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): 53.42 -> 53.99Improved
use both image-type and question-type negatives rather than one type (ablation)ConClaT · VQA-RephrasingsConsensus Score CS(4) (%): 53.77 -> 53.99Improved
raise joint-loss weight beta from 0.5 to 0.9 (ablation)Joint L_SSCL+L_CE · VQA-RephrasingsConsensus Score CS(4) (%): 53.63 -> 40.68Hurt

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