Please first install TensorFlow by following the official instructions. (Do not choose a preview release version of TensorFlow 2.0.) For GPU support, it is very important to install the specific versions of CUDA and cuDNN that are compatible with the respective version of TensorFlow.


We strongly recommend to use TensorFlow on a Linux or Windows system with a modern CUDA-enabled GPU from Nvidia, since TensorFlow currently does not support GPUs on the Mac. Without a GPU, training and prediction will be much slower (~30-60 times, even when using a computer with 40 CPU cores).

Second, we suggest to install Jupyter to be able to run our provided example notebooks that contain step-by-step instructions on how to use this package.

Finally, install the latest stable version of the CSBDeep package with pip:

pip install csbdeep


  • The package is compatible with Python 2 and 3, but mainly developed and tested with Python 3 (which we recommend to use).
  • If you use Python 3, you may need to use pip3 instead of pip.

Alternative GPU-enabled Installation (Linux only)

Since installing TensorFlow with its dependencies (CUDA, cuDNN) can be challenging, we offer a ready-to-use Docker container as an alternative to get started more quickly. (What is Docker?)