Readme.md

GANDLF Models

Current Models

Architecture	Link
Brain Age Predictor	https://doi.org/10.1093/brain/awaa160
Deep UNet	https://doi.org/10.1109/JSTARS.2018.2833382
DenseNet	https://arxiv.org/abs/2004.04968
EfficientNet	https://arxiv.org/abs/1905.11946
Fully Connected Network	https://arxiv.org/abs/1411.4038
ResNet	https://arxiv.org/pdf/1512.03385.pdf
SDNet	https://doi.org/10.1016/j.media.2019.101535
UInc	https://arxiv.org/abs/1907.02110
UNet	https://arxiv.org/abs/1606.06650v1
VGG	https://arxiv.org/abs/1409.1556

Adding a new model

• Follow example of GANDLF.models.unet.
• Define a new submodule under GANDLF.models, and define a class that inherits from GANDLF.models.ModelBase.
• Ensure that a forward pass is implemented.
• All parameters should be taken as input, with special parameters (for e.g., residualConnections for unet) should not be exposed to the parameters dict, and should be handled separately via another class.
  • For example, GANDLF.models.unet.unet has a residualConnections parameter, which is not exposed to the parameters dict, and a separate class GANDLF.models.unet.resunet is defined which enables this flag.
• Add the model's identifier to GANDLF.models.__init__.global_model_dict as appropriate.
• Call the new mode from the config using the model key.
• To update the tests, append the new model key to either in all_models_segmentation, all_models_segmentation or all_models_classification in test_full.py