Item-FindAR computer vision backend.
- Requires Python 3.8 or later.
- Run
pip install -r requirements.txt - Place trained model in
./models/. Certificates are included in./certs/for demo purposes. For future research or deployment you should regenerate them. - Run
python.exe main.py
Running the server on a local device with high-end graphics hardware and then connecting over a forward-proxy proved to be the easiest and cost-effective method for running the computer vision backend. Localtunnel is a free alternative to NGROK that allows for static domains, but uptime was patchy when we were testing. Ngrok requires a premium plan to register a domain, if using the free plan, remember to change the url to the one returned by the command below in \src\features\augment_system\workers\sendImg.js for the front-end before compiling with webpack: Front-end Repo.
- Install NGROK.
- Run the following command to start the proxy:
ngrok http https://localhost:8000 --domain ai.itemfindar.net
To check what frames the server is receiving, uncomment the line img.save("./saved-images/file-" + time.strftime("%Y%m%d-%H%M%S") + ".png") in main.py. Frames are stored in ./saved-images.
If "Queue Full" is displayed, restart the server. The error message when you stop the server may provide some information to explain the cause.