Installation

HOCT requires Python 3.11 or later.

Command line

With uv installed, run the CLI in an isolated environment:

uvx --from 'hoct[bioio]' hoct track images.tif segmentation.tif -o tracks.geff

For an existing Python environment:

pip install 'hoct[bioio]'
hoct --help

Python and optional dependencies

pip install hoct

Extra	Purpose
bioio	Image-file support for the tracking CLI
correction	Sparse-label fine tuning
demo	Correction dependencies, Napari with Qt, and Click for the demo
dev	Tests, linting, and development dependencies

The demo script lives in the repository; use a checkout to run it. See the incremental correction guide.

Tracking solver

HOCT uses tracksdata's ILP solver. Gurobi is included as a dependency and needs a valid licence. The solver attempts SCIP as a fallback when Gurobi is unavailable; the fallback must be available in your environment. A successful package installation alone does not guarantee that a particular solver is usable.

Models and devices

load_model() downloads general_v1 by default, verifies its SHA256, and caches it. Available names include general_v1, ctc_v0, and general_v0; use hoct.available_models() to query the registry. A local TorchScript .pt path also works. Set HOCT_CACHE_DIR to choose the download directory.

The CLI tries CUDA and falls back when it is unavailable. Choose --device cpu or --device mps explicitly if appropriate. In Python, pass an available device to load_model(device=...); the Python loader does not automatically select one.