Model cascade with Edge Impulse (FOMO-AD -> LLM)

This repository contains a demo using FOMO-AD (Edge Impulse's visual anomaly detection model) to find interesting parts of an image, then cascade to GPT4-o to do further analysis.

Prerequisites

You'll need a trained visual anomaly detection model in Edge Impulse. See the docs. Afterwards, download your model in .eim format, including hardware optimization via:

$ edge-impulse-linux-runner --clean --download path-to-your-model.eim

(Alternatively, go to Deployment in your Edge Impulse project, find the 'Linux' deployment for your architecture (e.g. 'Linux (AARCH64)'), and build from there).

Setup

1. Install dependencies:

   npm install
   

2. Set your OpenAI API Key:

   export OPENAI_API_KEY=sk-MA...
   

3. Run the application:

   npm run build && node build/classify-camera-webserver.js ./path-to-your-model.eim
   

   If you have multiple cameras a message will be printed, and you should add the camera name as the last argument to the script above.

4. Go to http://localhost:4912/ and see the cascade working.