StreetSight: Auditing Public Spaces for Missing Women and Urban Infrastructure Using Google Street View

Analyzing gender representation and infrastructure quality in Mumbai, Delhi, and Navi Mumbai using Google Street View imagery.

Key Findings

Proportion Women

City	Women (avg)	Men (avg)	Prop. Women	N Locations
Mumbai	0.73	3.12	0.21	498
Delhi	0.38	1.58	0.21	500
Navi Mumbai	0.48	1.49	0.24	437

Women are significantly underrepresented in public spaces across all three cities, comprising only 21-24% of visible pedestrians.

By Road Type

Road Type	Prop. Women	N Annotations
Primary	0.11	79
Secondary	0.17	166
Tertiary	0.19	293
Residential	0.24	1,362

Proportion of women is lowest on primary/secondary roads and highest in residential areas.

Infrastructure

City	Potholes	Litter	Footpath	Lane Markings
Delhi	1%	27%	19%	17%
Mumbai	1%	21%	50%	29%
Navi Mumbai	1%	28%	29%	23%

Data Pipeline

scripts/01_sample_locations.py    Sample random road segments from OSM
         ↓
scripts/02_check_coverage.py      Check Street View coverage (free API)
         ↓
scripts/03_download_images.py     Download images (paid API, ~$7/1000 images)
         ↓
scripts/create_labelstudio_tasks.py    Generate Label Studio import file
         ↓
Label Studio                      Human annotation of images
         ↓
notebooks/01_pipeline.ipynb       Pipeline quality & bias assessment
         ↓
notebooks/02_annotations.ipynb    Analysis and visualization

Setup

# Clone and install
git clone https://github.com/soodoku/missing_women_gsview
cd missing_women_gsview
uv sync

# Set up API key
cp .env.example .env
# Edit .env with your Google Street View API key

Usage

# 1. Sample locations from OpenStreetMap
uv run python scripts/01_sample_locations.py

# 2. Check Street View coverage
uv run python scripts/02_check_coverage.py

# 3. Download images (requires API key, costs money)
uv run python scripts/03_download_images.py

# 4. Create Label Studio tasks
uv run python scripts/create_labelstudio_tasks.py

# 5. Run analysis notebooks
uv run jupyter notebook notebooks/01_pipeline.ipynb
uv run jupyter notebook notebooks/02_annotations.ipynb

Data

• Sampled locations: 7,000 road segments (2,500 Mumbai, 2,500 Delhi, 2,000 Navi Mumbai)
• Annotated images: 1,942 tasks with human annotations
• Annotation fields: women_count, men_count, potholes, litter, footpath, lane_markings, land_use

Directory Structure

data/
  annotations/       Raw Label Studio exports
  coverage/          Street View coverage results
  images/            Downloaded Street View images
  samples/           Sampled road locations
  roads/             OSM road data
  labelstudio_tasks.json
notebooks/
  01_pipeline.ipynb  Pipeline quality & bias assessment
  02_annotations.ipynb  Annotation analysis
  outputs/           Generated outputs (CSVs, PNGs, HTMLs)

Output Files

File	Description
notebooks/outputs/combined_annotations.csv	Merged annotation dataset
notebooks/outputs/city_summary.csv	Summary statistics by city
notebooks/outputs/annotated_locations_map.html	Interactive map
notebooks/outputs/sex_ratio_heatmap.html	Prop. women heatmap

License

MIT