AIROGS Challenge Algorithm

This is the github repository for the Schulich Applied Computing in Medicine's submission to the AIROGS challenge. It builds a Docker Container that works with the Grand Challenge website.

Modifying pretrained networks

A dropout layer was inserted before the last fully connected layer in all networks used. As an example, this is how it is done for Densenet. The full code for this is available in process.py. If you want to load the trained weights outside of this docker container, you will need to modify the pretrained model appropriately.

model = models.densenet161(pretrained=False)
num_ftrs = model.classifier.in_features
model.classifier = nn.Sequential(nn.Dropout(dropout), nn.Linear(num_ftrs, 1, bias=True))
model.load_state_dict(torch.load(DENSENET_WEIGHT_PATH))

Building on Windows

If you want to use this container on Windows, you will need to have Docker installed on your system with WSL 2.0. It should be possible to use CUDA with WSL, but this was not tested. If no GPU is detected, Pytorch will use CPU.

Competition Paper

The competition paper will be uploaded when the challenge ends.