Mixed-scale Dense Networks for PyTorch

An implementation of Mixed-Scale Dense networks in PyTorch.

This paragraph should contain a high-level description of the package, with a brief overview of its features and limitations.

  • Free software: GNU General Public License v3

  • Documentation: [https://ahendriksen.github.io/msd_pytorch]

Getting Started

It takes a few steps to setup Mixed-scale Dense Networks for PyTorch on your machine. We recommend installing Anaconda package manager for Python 3.

Requirements

This package requires

This package is compatible with python 3.6 and 3.7.

Installing with Conda

Simply install with either cuda version 9.0 or version 10.0:

conda install -c conda-forge -c aahendriksen -c pytorch msd_pytorch cudatoolkit=9.0
# or
conda install -c conda-forge -c aahendriksen -c pytorch msd_pytorch cudatoolkit=10.0

Installing from source

To install msd_pytorch from source, you need to have the CUDA toolkit installed. Specifically, you need nvcc and a compatible C++ compiler. Moreover, you need to have a working installation of PyTorch.

To install PyTorch, we recommend using conda. Install PyTorch with either of these versions of cudatoolkit:

conda install pytorch=1.1.0 torchvision cudatoolkit=10.0 -c pytorch
conda install pytorch=1.1.0 torchvision cudatoolkit=9.0  -c pytorch

To get the source code, simply clone this GitHub project.

git clone https://github.com/ahendriksen/msd_pytorch.git
cd msd_pytorch

Using pip to install the package automatically triggers the compilation of the native C++ and CUDA code. So you need to direct the installer to a CUDA-compatible C++ compiler in this way:

GXX=/path/to/compatible/cpp/compiler pip install -e .[dev]

Or, if the standard C++ compiler is compatible with CUDA:

pip install -e .[dev]

Using the tools

The msd_pytorch package ships with some command-line tools to make your life easier. If you have input and target images in directories ./train/input/ and ./train/target/, then you can train a network to do regression with the following command in your terminal:

msd regression -p with train_input_glob='./train/input/*' train_target_glob='./train/target/*' epochs=10 msd.depth=30

Similarly, segmentation is possible using the following command:

msd segmentation -p with train_input_glob='./train/input/*' train_target_glob='./train/target/*' epochs=10 msd.depth=30 labels=[0,1,2,3]

More command-line arguments are available

epochs = 1                         # The number of epochs to train for
labels = [0, 1]                    # The labels that you expect in your segmentation targets (if you are doing segmentation)
train_input_glob = ''              # The glob pattern for the training set input data
train_target_glob = ''             # The glob pattern for the training set target data
val_input_glob = ''                # The glob pattern for the validation set input data
val_target_glob = ''               # The glob pattern for the validation set input data
msd:
  c_in = 1                         # Number of input channels
  c_out = 1                        # Number of output channels (for regression; see `labels` for segmentation)
  depth = 10                       # The depth of the network
  width = 1                        # The width of the network
  dilations = [1, 2, ..., 10]      # The dilation-scheme that is used in the network
  loss = 'L2'                      # Which loss to use for regression (options: "L1" or "L2")

The path specification for the images is a path with optional glob pattern describing the image file paths. Tildes and other HOME directory specifications are expanded with os.path.expanduser and symlinks are resolved.

If the path points to a directory, then all files in the directory are included. If the path points to file, then that single file is included.

Alternatively, one may specify a “glob pattern” to match specific files in the directory.

Examples:

  • "~/train_images/"

  • "~/train_images/cats*.png"

  • "~/train_images/*.tif"

  • "~/train_images/scan*"

  • "~/train_images/just_one_image.jpeg"

Running the examples

To learn more about the functionality of the package check out our examples folder.

Authors and contributors

  • Allard Hendriksen - Initial work

  • Jonas Adler - Discussions and code

See also the list of contributors who participated in this project.

How to contribute

Contributions are always welcome. Please submit pull requests against the master branch.

If you have any issues, questions, or remarks, then please open an issue on GitHub.

License

This project is licensed under the GNU General Public License v3 - see the LICENSE.md file for details.