/deepcell-applications

Run DeepCell Applications

Primary LanguagePythonOtherNOASSERTION

deepcell-applications

Build Status Coverage Status Apache 2.0

A script and runnable Docker image for plugging DeepCell Applications (like Mesmer) into existing pipelines.

Running the Python script

The run_app.py script is used to read the input files from the user and process them with the selected Application. An example Python script run_app.py is provided as an example deepcell.applications workflow.

Script arguments

The first required argument to the script is the Application name: python run_app.py APP_NAME. Each supported application has a variety of different configuration arguments. Below is a table summarizing the currently supported applications and their arguments and any defaults. For more information, use python run_app.py --help or python run_app.py APP_NAME --help.

To learn more about the pretrained models, see the introductory documentation.

Mesmer arguments

Name Description Default Value
--output-directory Directory to save output file. "./output"
--output-name The name for the output file. "mask.tif"
--nuclear-image REQUIRED: The path to an image containing the nuclear marker(s). ""
--nuclear-channel The numerical index of the channel(s) from nuclear-image to select. If multiple values are passed, the channels will be summed. 0
--membrane-image The path to an image containing the membrane marker(s). If not passed, an array of zeroes will be used instead. ""
--membrane-channel The numerical index of the channel(s) from membrane-image to select. If multiple values are passed, the channels will be summed. 0
--compartment Predict nuclear or whole-cell segmentation. "whole-cell"
--image-mpp The resolution of the image in microns-per-pixel. A value of 0.5 corresponds to 20x zoom. 0.5
--batch-size Number of images to predict on per batch. 4
--squeeze Whether to np.squeeze the outputs before saving as a tiff. False

Script command

export DATA_DIR=/Users/Will/vanvalenlab/example_data/multiplex
export APPLICATION=mesmer
export NUCLEAR_FILE=example_nuclear_image.tif
export MEMBRANE_FILE=example_membrane_image.tif
python run_app.py $APPLICATION \
  --nuclear-image $DATA_DIR/$NUCLEAR_FILE \
  --nuclear-channel 0 \
  --membrane-image $DATA_DIR/$MEMBRANE_FILE \
  --membrane-channel 0 1 \
  --output-directory $DATA_DIR \
  --output-name mask.tif \
  --compartment whole-cell

Using Docker

The script can also be run as a Docker image for improved portability. This repository has published versions for both CPU and GPU for each versioned release.

The latest images for each are readily available:

  • vanvalenlab/deepcell-applications:latest (CPU build)
  • vanvalenlab/deepcell-applications:latest-gpu (GPU build)

Or for better reproducibility, use a versioned release (e.g. vanvalenlab/deepcell-applications:0.1.0-gpu)

Run the image

For Docker API version >= 1.40:

export DATA_DIR=/path/to/data/dir
export MOUNT_DIR=/data
export APPLICATION=mesmer
export NUCLEAR_FILE=example_nuclear_image.tif
export MEMBRANE_FILE=example_membrane_image.tif
docker run -it --gpus 1 \
  -v $DATA_DIR:$MOUNT_DIR \
  vanvalenlab/deepcell-applications:latest-gpu \
  $APPLICATION \
  --nuclear-image $MOUNT_DIR/$NUCLEAR_FILE \
  --membrane-image $MOUNT_DIR/$MEMBRANE_FILE \
  --output-directory $MOUNT_DIR \
  --output-name mask.tif \
  --compartment whole-cell

For Docker API version < 1.40:

export DATA_DIR=/path/to/data/dir
export MOUNT_DIR=/data
export APPLICATION=mesmer
export NUCLEAR_FILE=example_nuclear_image.tif
export MEMBRANE_FILE=example_membrane_image.tif
docker run -it \
  -v $DATA_DIR:$MOUNT_DIR \
  vanvalenlab/deepcell-applications:latest-gpu \
  $APPLICATION \
  --nuclear-image $MOUNT_DIR/$NUCLEAR_FILE \
  --membrane-image $MOUNT_DIR/$MEMBRANE_FILE \
  --output-directory $MOUNT_DIR \
  --output-name mask.tif \
  --compartment whole-cell

Build the image

It is also very easy to build a custom image to test any new functionality:

docker build -t vanvalenlab/deepcell-applications .

It is also possible to change the base DeepCell version when building the image, using the build-arg DEEPCELL_VERSION. This makes it simple to build a CPU-only version of the image or to build a new version of deepcell-tf.

# the -gpu tag is required to enable GPU compatibility when overriding versions
docker build --build-arg DEEPCELL_VERSION=0.9.0-gpu -t vanvalenlab/deepcell-applications .