highresnet 0.4.0

PyTorch implementation of HighResNet

HighRes3DNet

PyTorch implementation of HighRes3DNet from Li et al. 2017, On the Compactness, Efficiency, and Representation of 3D Convolutional Networks: Brain Parcellation as a Pretext Task.

All the information about how the weights were ported from NiftyNet can be found in my submission to the MICCAI Educational Challenge 2019.

A 2D version (HighRes2DNet) is also available.

Installation

1. Install PyTorch

Within a conda environment:

$ conda create -n deepgif python -y
$ conda activate deepgif
(deepgif) $ conda install pytorch torchvision cudatoolkit=10.0 -c pytorch -y

2. Install the pip package

(deepgif) $ pip install highresnet

>>> from highresnet import HighRes3DNet
>>> model = HighRes3DNet(in_channels=1, out_channels=160)

Usage

PyTorch Hub

If you are using pytorch>=1.2.0, you can import the model directly from this repository using PyTorch Hub.

>>> import torch
>>> repo = 'fepegar/highresnet'
>>> model_name = 'highres3dnet'
>>> print(torch.hub.help(repo, model_name))

        "HighRes3DNet by Li et al. 2017 for T1-MRI brain parcellation"
        "pretrained (bool): load parameters from pretrained model"

>>> model = torch.hub.load(repo, model_name, pretrained=True)

Command line interface

(deepgif) $ deepgif t1_mri.nii.gz parcellation.nii.gz
Using cache found in /home/fernando/.cache/torch/hub/fepegar_highresnet_master
100%|███████████████████████████████████████████| 36/36 [01:13<00:00,  2.05s/it]