Bridging the Digital Divide: Performance Variation across SocioEconomic Factors in Vision-Language Models

This repo contains code (in Jupyter Notebooks) for the analysis featured in our EMNLP 2023 paper:

($ equal contribution)

Data

The data we used is from the GapMinder project called Dollar Street.

View the images of various households across the world through Dollar Street: photos as data to kill country stereotypes

The dataset is available for download at ML Commons Dollar Street Dataset

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
.ipynb_checkpoints		.ipynb_checkpoints
CLIP_other_versions		CLIP_other_versions
Images		Images
data		data
01- Process Images and Run CLIP.ipynb		01- Process Images and Run CLIP.ipynb
02 - Merge CLIP results.ipynb		02 - Merge CLIP results.ipynb
03-RQ1 Analysis Plots.ipynb		03-RQ1 Analysis Plots.ipynb
03.1 - RQ1_ Image exploration by CLIP Score.ipynb		03.1 - RQ1_ Image exploration by CLIP Score.ipynb
04- RQ2 Run CLIP (seperated topics).ipynb		04- RQ2 Run CLIP (seperated topics).ipynb
05- Analysis_RQ2.ipynb		05- Analysis_RQ2.ipynb
06- Get CLIP_results_sep.ipynb		06- Get CLIP_results_sep.ipynb
07- Analysis_RQ3.ipynb		07- Analysis_RQ3.ipynb
README.md		README.md