AIDA is a crisis intelligence platform that uses LLMs to quickly detect disasters, collect data to help further understand the nature of the challenge, extract actionable insights and empower humanitarian organisations to respond effectively.
The main model used in Aida is Llama 3.2. It is a 11B model that is able to understand satellite, drone and ground images and detect accurately whether there is a natural disaster happening.
It is being used to perform the following tasks
- Detect the presence of a disaster in before and after satellite images of a region
- Extract the aid resources required by victims of a disaster
- Describe images of a disaster
- Describe the current state of a disaster in real-time using commentary and images
We also get top use the SAM2 model to segment the disaster area from the image for further analysis.
These features make it possible to achieve the powerful features of the AIDA platform.
Presentation: https://tinyurl.com/3xvw2vn6
-
Clone the repo and navigate to the root folder.
-
To run the app using Docker, make sure you've got Docker installed on your system. From the project's root directory, run:
make run-container
If you want to run the app locally, without using Docker, then:
-
Clone the repo and navigate to the root folder.
-
Install uv for dependency management.
-
Start the app. Run:
make run-local
This will set up a virtual environment
.venvin the current directory with Python 3.11, install dependencies, and start the Uvicorn server.
- Priya Prakash
- Nitish Mital
- Nihir Vedd
- Habeeb Shopeju