Model and Datasets

DeepCell models and training datasets are licensed under a modified Apache license for non-commercial academic use only. An API key for accessing datasets and models can be obtained at https://users.deepcell.org/login/.

API Key Usage

The token that is issued by https://users.deepcell.org should be added as an environment variable:

export DEEPCELL_ACCESS_TOKEN=<token-from-users.deepcell.org>

This line can be added to your shell configuration (e.g. .bashrc, .zshrc, .bash_profile, etc.) to automatically grant access to DeepCell models/data upon login.

Models

The model can be downloaded for local use:

>>> from deepcell_types.utils import download_model

>>> download_model()

A specific version can be requested:

download_model(version="2026-06-15")

A listing of available pre-trained model versions is available from deepcell_types.utils.list_model_versions().

To fetch a baseline checkpoint instead of the main DeepCellTypes model:

from deepcell_types.utils import download_baseline_checkpoint

# One of: "cellsighter", "maps", "xgboost"
download_baseline_checkpoint("maps")

All three baselines ship in a single compressed bundle, which holds each one’s weights plus any companion file needed at inference (maps ships _stats.npz; xgboost ships .remap.json). The bundle is unpacked automatically and the returned list gives the local path of every file for the baseline you asked for.

Because the bundle is shared, the first call downloads all three baselines (646 MB) regardless of which one you request; the other two are then served from the local cache without a further download.

The Nimbus baseline is not served here: its pretrained weights are distributed upstream, so install it with pip install nimbus-inference==0.0.5 on Python 3.11 (which fetches the weights automatically) rather than download_baseline_checkpoint.

Training Data

Warning

The training dataset is over 1.3 TB - make sure you have space and sufficient network bandwidth before attempting to download.

Similarly, training data can be downloaded for local use with:

>>> from deepcell_types.utils import download_training_data
>>> download_training_data()