Tool for applying several deep learning methods on the MRNet and KneeXray dataset.
Current implementations:
-
DIAGNOSIS (MRNet based)
-
DCGAN
-
VQVAE - VQVAE2
-
INTROVAE
-
BIGAN
Own Creations:
- MoCoAE, MocoGAN
- Transition Networks for pseudo 3D analysis for DCGAN, INTROVAE, BIGAN
Later this will simply be something like:
pip install deeptool
For now the source code needs to be downloaded. Then navigation to the downloaded folder is required and the package can be installed with:
pip install -e .
from deeptool.train_loop import get_model, test_one_batch, main_loop
from deeptool.parameters import get_all_args, compat_args
check parameters.py for more information about the effect of each parameter
# Define Arguments (Example)
args = get_all_args()
args.dim = 2 # Dimension of network reduced to 2
args.batch_size = 1 # How many pictures are included per training update
args.n_res_2d = 1 # architecture contains resnet blocks
args.watch_batch = 100 # visualize every 100 batches
args.model_type = "introvae" # train the introvae model
args = compat_args(args) # solve argument interactions
main_loop(args) # run the training
Watch the model improve on the task, as displayed below...