AI Paper Atlas

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

WinoGAViL: Gamified Association Benchmark to Challenge Vision-and-Language Models

Yonatan Bitton, Nitzan Bitton Guetta, Ron Yosef, Yuval Elovici, Mohit Bansal, Gabriel Stanovsky, Roy Schwartz

NeurIPS 2022 · Vision-language and multimodal › Visual question answering · approach: Visual reasoning diagnostics · 6 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
fine-tune with supervised binary classifierCLIP ViT-B/32 · WinoGAViL game, 10&12 candidatesJaccard score: 42 -> 49Improved
fine-tune with supervised binary classifierCLIP ViT-B/32 · WinoGAViL game, 5&6 candidatesJaccard score: 53 -> 52No clear effect
increase number of candidates from 5&6 to 10&12CLIP-ViL · WinoGAViL gameJaccard score: 47 -> 15Hurt
increase number of candidates from 5&6 to 10&12CLIP-RN50 · WinoGAViL gameJaccard score: 50 -> 35Hurt
increase number of candidates from 5&6 to 10&12Humans · WinoGAViL gameJaccard score: 92 -> 90No clear effect
use game-collected associations instead of SWOW-basedCLIP-VIT-L/14 · 5&6 candidatesJaccard score: 74 -> 53Hurt
replace image candidates with OFA captions and text modelMPNet QA · WinoGAViL game 10&12Jaccard score: 47 -> 52Mixed
use image captions with MPNet QA instead of MPNettext models · WinoGAViL game 10&12Jaccard score: 39 -> 47Improved
visually non-salient cue instead of visually salient cueCLIP ViT-B/32 · WinoGAViL game 10&12model accuracy: 75 -> 36Hurt
make both cue and candidates images instead of text (SWOW)CLIP-ViT-L/14 · SWOW modalitiesJaccard index: 86 -> 65Hurt
CLIP-RN50 with 3 associations vs 2 for 12 candidatesCLIP-RN50 · WinoGAViL gamemodel performance: 37 -> 33Hurt

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