AI Paper Atlas

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

Grounding Answers for Visual Questions Asked by Visually Impaired People

Chongyan Chen, Samreen Anjum, Danna Gurari

CVPR 2022 · Vision-language and multimodal › Visual question answering · approach: Multimodal question answering datasets · 50 citations (OpenAlex)

Check 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
Train MAC-Caps on VizWiz-VQA rather than GQA, CLEVR, or VQA-v2MAC-Caps · VizWiz-VQA-Grounding testAvg IoU: 12.56 -> 27.43; 15.31 -> 27.43; 17.42 -> 27.43Improved
Train LXMERT on VizWiz-VQA rather than OSCAR on VQA-v2VQA grounding models · VizWiz-VQA-Grounding testAvg IoU: 15.48 -> 22.09Improved
Use MAC-Caps trained on VizWiz-VQA rather than LXMERT trained on VizWiz-VQAVizWiz-pretrained grounding models · VizWiz-VQA-Grounding testAvg IoU: 22.09 -> 27.43Improved
Predict the whole image as the grounded area instead of using a learned grounding modelVizWiz-VQA-Grounding · VizWiz-VQA-Grounding testAvg IoU: 27.43 -> 33%Improved
Evaluate only questions whose answers the model predicted correctlyMAC-Caps pretrained on VizWiz-VQA · VizWiz-VQA-Grounding testAvg IoU: 27.43 -> 32.8Improved
Ground small answer regions rather than medium or large onesLXMERT, OSCAR, and MAC-Caps pretrained on VizWiz · VizWiz-VQA-Grounding test, small image coverageIoUHurt
Evaluate on images with no quality issues rather than images with medium or severe issuesLXMERT and OSCAR · VizWiz-VQA-Grounding test, image qualityIoU: 0.28, 0.47 and 0.83Hurt
Ground questions requiring text recognition or counting rather than object or color recognitionLXMERT, OSCAR, and MAC-Caps pretrained on VizWiz · VizWiz-VQA-Grounding test, vision skillsIoUHurt
Pretrain grounding models on VizWiz-VQA rather than contrived VQA datasetsVQA and answer-grounding models · VizWiz-VQA-Grounding testmAP: 0.01% -> 0.96%; 0.02% -> 0.96%; 0.03% -> 0.96%; 0.08% -> 0.96%Improved

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