GeoTagr 🌍 📸 🤖

GeoTagr is an AI-powered tool that predicts the geographical location of photos using a LangChain-based agent system powered by Google's Gemini Flash 1.5 LLM.

Overview 🎯

GeoTagr analyzes images and returns predicted coordinates along with detailed reasoning for its predictions. You can test it HERE

Project Structure 📁

.
├── src/          # Main application code and agent implementation
└── test/         # Test images and benchmarking scripts

Usage 🔄

Prerequisites ⚙️

1. Clone the repository:

   git clone https://github.com/mwitjez/GeoTagrBacked.git
   cd GeoTagrBacked

2. Set up environment variables:

   # Copy the template
   cp .env_template .env
   
   # Edit .env with your credentials
   nano .env

3. Install dependencies:

   pip install -r requirements.txt

Option 1: REST API 🌐

1. Start the API server:

   uvicorn api:app --reload --port 8000

2. Access the API:

   • Interactive Documentation: Visit http://localhost:8000/docs
   • Direct API endpoint: http://localhost:8000/process-image

3. Make a prediction:

   curl -X POST "http://localhost:8000/process-image" \
     -H "Content-Type: multipart/form-data" \
     -F "file=@/path/to/your/image.jpg"

4. Response Format:

   {
     "latitude": str,
     "longitude": str,
     "reasoning": str,
   }

Option 2: Command Line 💻

1. Run with a single image specified in main.py:

   python src/main.py

Option 3: Streamlit 🎈

1. Run using Streamlit interface:

   streamlit run app.py

Performance 📊

Benchmark results on the Im2GPS dataset:

Metric	Accuracy
Within 1km	13.9%
Within 25km	43.9%
Within 200km	64.1%
Within 750km	80.2%
Within 2500km	92.8%

Additional Statistics:

• Mean Distance Error: 843.23 km
• Median Distance Error: 49.06 km