bearvision

A frontend and backend for automatic scouting of FRC

Set up environment

Using pip:

$ python3 -m venv .venv
$ source .venv/bin/activate
$ pip install opencv-python torch torchvision streamlit b2sum
$ mkdir -p samples/scores/{data,images}

Using uv:

$ uv venv .
$ source .venv/bin/activate
$ uv pip install opencv-python torch torchvision streamlit b2sum
$ mkdir -p samples/scores/{data,images}

Getting training data

It is recommended to use yt-dlp to download videos. Some examples are provided in vids.txt:

$ yt-dlp -tmp4 -a vids.txt -o "./samples/%(title)s.%(ext)s"

Because YouTube is constantly changing their API, yt-dlp may fail if using a packaged version. Get the nightly version here.

Run annotator

$ streamlit run annotation/streamlit_app.py

Run trainer

$ python3 training/trainer.py