ManualCellLabeler

Interactive Streamlit app for manual neuron label confirmation using NeuroPAL and related HDF5/CSV data.

Usage

• Run the Streamlit app locally:

source .venv/bin/activate && streamlit run app.py

• Open the UI: visit http://localhost:<port_number> (Firefox is the recommended browser for this app)

Notes

• Default data root used by the app: /store1/shared/flv_utils_data/ (see FLV_UTILS_DATA_DIR in app.py).
• The app expects project subfolders with autolabel_data, neuron_rois, processed_h5, and labels_csv.

Development

• Python >= 3.10. Use the included .venv for reproducible environment.
• Dependencies are listed in pyproject.toml.

Containerized deployment

Use run.sh to build (if needed) and launch the container:

./run.sh /store1/path/to/your/data_dir

Optional second argument for a custom output directory (defaults to <data_dir>/relabelled):

./run.sh /store1/path/to/your/data_dir /store1/path/to/output_dir

A free port is auto-selected from 3001-3030, so multiple users can run simultaneously. In case you would like to force a specific port (e.g. 3005):

PORT=3005 ./run.sh /store1/path/to/your/data_dir

The assigned port is printed on startup (Firefox recommended).

The image is built automatically on first run or after a reboot (~35s cold build). Podman storage is kept under /tmp/<user>-podman-storage/ to work around NFS xattr limitations.

Integrate relabels to labels_csv on CLI

uv run python update_labels_from_log.py /store1/path/to/output_dir/$(neuron_class)_$(user)_log.csv --apply