dvigneault/dvpy

Install:

$ pip install git+https://github.com/dvigneault/dvpy.git#egg=dvpy

The only dependency not installed by default is tensorflow, as we do not want to specify the GPU or CPU version. These can be installed as follows:

$ pipenv install tensorflow # CPU
$ pipenv install tensorflow-gpu # GPU

Run tests:

First, build your development environment (requires that Docker be installed and running):

$ ./docker-build.sh

Spin up your container:

$ ./docker-run.sh

And run your tests!

$ pytest                                      # Run all the tests
$ pytest ./test/test_find_duplicates.py       # Run a specific test
$ pytest ./test/test_find_duplicates.py -v    # Be verbose
$ pytest ./test/test_find_duplicates.py -v -s # Be super verbose

dvpy's Issues

UNet: Take arbitrary input sizes.

Currently, inputs to the unet are restricted in terms of their dimensions, due to the fact that successive downsampling/upsampling must not change the dimension (i.e., dim / 2**depth * 2**depth == dim). This could be solved by padding the result of UpSampling to match the layer with which it is being concatenated. It would be important to ensure that no translations occur, so this must be unit tested.