GPU accelerated image registration

This package provides GPU accelerated image registration. It will fall back to CPU computation if GPU is not available.

For example, given this input:

This is the aligned image output:

How to run the program

First, create the python environment if you do not have one set up already.

$ cd imagereg
$ conda create -n imagereg python=3.6.8 pip
$ source activate imagereg
$ pip install -r requirements.txt
$ pip install https://github.com/DeMarcoLab/imagereg.git

Then there are two options for running the program: from the command line, and from within python.

The pipeline image registration function takes three input arguments:

The path to the input image directory
The regex string describing the file naming. Filenames matching this regex are sorted alphabetically before image alignment.
The path to an empty output directory where you would like to save the results. This must be an empty directory.

1. From the command line

conda activate imagereg
mkdir output_directory
python path/to/imagereg/main.py tests/images img[0-9].tif output_directory

2. From within python

$ conda activate imagereg
$ mkdir output_directory
$ python
>>> import imagereg
>>> imagereg.pipeline('tests/images', 'img[0-9].tif', 'output_directory')

Setting up your development environment

Create the conda development environment

conda create -n imagereg python=3.6.8 pip
source activate imagereg
pip install -r requirements/requirements-default.txt
pip install -r requirements/requirements-dev.txt

Optional GPU setup, for accelerated performance. You must have a supported NVIDIA GPU.

i. Install CUDA version 10.0 (see https://docs.nvidia.com/cuda/)

ii. Install cupy into your imagereg conda environment:
```
source activate imagereg
pip install cupy==5.4.0
```
Fork and clone the repository at https://github.com/DeMarcoLab/imagereg.git
Install imagereg into your environment as an editable installation:
```
cd imagereg
pip install -e .
```

Running the tests

We use pytest for our unit testing.

cd imagereg
pytest

DeMarcoLab / imagereg