gui

SST-GUI

This repo is a GUI (via Streamlit) for the idea presented in SST.

1. Run uv run streamlit run gui.py. This starts a Streamlit app (which you can open in a browser) to specify where your images are located, speciy how to group images, and pick out reference frames and the corresponding masks. All of this data will be saved in a spec.json file.
2. Run uv run inference.py spec.json. The spec.json file will include all of the decisions you made with the GUI, and will then generate masks for the rest of the images using SAM 2 and the reference masks you generated with the GUI.

GUI

First, upload metadata and tell the system how to get your images.

Then wait for SAM 2 to get all masks (with no prompt) from 5 images.

Then filter the masks to the objects you care about and label the objects with integer ids.

Finally, save your masks to disk. Then run uv run inference.py spec.json.

You can validate that your images work by using uv run view.py PATH: