https://github.com/apsurt/omni-geo-ai
Omni Geoguessr AI: A Vision Transformer AI integrated with Geoguessr for automated geographic location prediction and gameplay using streetview panoramas.
https://github.com/apsurt/omni-geo-ai
ai automated-gameplay city-recognition climate-classification computer-vision deep-learning elevation-detection geography geoguessr geolocation google-maps-api image-recognition location-prediction machine-learning python streetview vision-transformer
Last synced: about 1 month ago
JSON representation
Omni Geoguessr AI: A Vision Transformer AI integrated with Geoguessr for automated geographic location prediction and gameplay using streetview panoramas.
- Host: GitHub
- URL: https://github.com/apsurt/omni-geo-ai
- Owner: Apsurt
- License: mit
- Created: 2024-05-30T07:25:33.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-07-21T19:28:32.000Z (about 1 year ago)
- Last Synced: 2024-07-21T21:14:03.095Z (about 1 year ago)
- Topics: ai, automated-gameplay, city-recognition, climate-classification, computer-vision, deep-learning, elevation-detection, geography, geoguessr, geolocation, google-maps-api, image-recognition, location-prediction, machine-learning, python, streetview, vision-transformer
- Language: Python
- Homepage:
- Size: 487 KB
- Stars: 1
- Watchers: 4
- Forks: 1
- Open Issues: 7
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- License: LICENSE
- Roadmap: ROADMAP.md
Awesome Lists containing this project
README
# Omni Geoguessr AI (Phase 1: Data & Hemisphere Model)
## Overview
This project focuses on building a pipeline to collect Google Street View panoramas and training a Vision Transformer (ViT) AI to predict the hemisphere (Northern/Southern) from images. This phase lays the groundwork for future geographic recognition models.
## Features
- **Automated Data Collection**: Fetches and labels panoramas using Google Street View API.
- **Hemisphere Prediction**: ViT-B/16 model trained to classify panoramas into Northern/Southern hemispheres.
- **Scalable Storage**: Images stored in Google Cloud Storage, metadata in PostgreSQL.
## Getting Started
### Prerequisites
- Python 3.10+
- PostgreSQL
- Google Cloud account with:
- Street View Static API enabled
- Geocoding API enabled
- Google Cloud Storage
### Installation
1. Clone the repository:
```bash
git clone https://github.com/Apsurt/apsurt-omni-geo-ai.git
cd apsurt-omni-geo-ai
```
2. Install dependencies:
```bash
pip install -r requirements.txt
```
3. Configure environment variables:
```bash
cp .env.example .env
# Add your Google API key and PostgreSQL credentials to .env
```
4. Initialize the database:
```bash
# Create PostgreSQL database
createdb omni_geo_ai
# Run the data collection script to initialize tables
python scripts/fetch_panoramas.py --num_images 5 --dry_run
```
### Usage
#### Data Collection
Run the panorama scraper to collect a specific number of images:
```bash
python scripts/fetch_panoramas.py --num_images 1000
```
Advanced options:
```bash
# Control the ratio of Northern/Southern hemisphere images
python scripts/fetch_panoramas.py --num_images 100 --north_ratio 0.7
# Save collection statistics to a JSON file
python scripts/fetch_panoramas.py --num_images 50 --output_stats stats.json
# Enable verbose logging
python scripts/fetch_panoramas.py --num_images 20 --verbose
# Clear existing data before collecting new images
python scripts/fetch_panoramas.py --num_images 50 --clear
# Only clear data without collecting new images
python scripts/fetch_panoramas.py --clear_only
# Update the StreetViewClient with ~1000 cities
python scripts/update_streetview_cities.py
```
#### Database Backup
Backup the PostgreSQL database to Google Cloud Storage:
```bash
python scripts/backup_db.py
```
Backup options:
```bash
# Specify a different GCS bucket
python scripts/backup_db.py --bucket my-backup-bucket
# Custom backup directory path in GCS
python scripts/backup_db.py --backup_dir database/daily
```
#### Train Hemisphere Model
```bash
python train.py --model vit_hemisphere --epochs 50
```
## License
MIT License. See [LICENSE](LICENSE).
## Roadmap
See [ROADMAP.md](ROADMAP.md) for phase details.
## Contact
tymon.becella@gmail.com