NOTE : Each of these folders below also contains README files to describe the contents of each folder.
The unet models in the hw5_layer, 4_layer, 6_layers and 5_layer_stride_4 folders use the DiceLoss function
The unet models in the BCE folder use the BCELogitLoss function
The best model is present in the "best_model" folder
|-----------------------------------------------------------------------------------------------------------------------------------------------|
Folder Name | Folder Description |
---|---|
hw5_layer | This folder contains the code for 5 layer Monai model used in Homework 5. This unet model |
has channels = (32, 64, 128, 256, 512) and strides = (2, 2, 2,2). Loss function used = DiceLoss The results and | |
graphs of this model are saved as png files | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
4_layer | This folder contains the code for 4 layer Monai model. This unet model has channels = (64, 128, 256, 512) |
and strides = (2, 2, 2). Loss function used = DiceLoss The results and graphs of this model are saved as png files. | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
6_layers | This folder contains the code for 6 layer Monai model. This unet model has channels = (32, 64, 64, 128, 256, 512) |
and strides = (2, 2, 2,2,2). Loss function used = DiceLoss The results and graphs of this model are saved as png files | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
5_layer_stride_4 | This folder contains the code for 5 layer Monai model where value of the strides are inccreased. This unet |
model has channels = (32, 64, 128, 256, 512) and strides = (4, 4, 4,4). Loss function used = DiceLoss. | |
The results and graphs of this model are saved as png files. | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
BCE (this has sub- | This folder contains the code of pytorch model (in the "pytorch" subfolder) and Monai model (in the "monai" subfolder) |
folders "pytorch" | Both the models have to same number of layers, however the pytorch model is created using the convolutional layers using |
and "monai" | pytorch library and the monai unet model is created using the convolutional blocks provided by monai. |
These models use the BCE loss function instead of dice loss | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
best_model | This folder contains the model files of the best performing models and also the ipymb files of the specific unet models |
being tested. the ipymb files are: | |
1) testing-monai-ADN-70-epochs : contains the monai unet model created using ADN Blocks trained for 70 epochs | |
2) testing-monai-ADN-200-epochs : contains the monai unet created using ADN Blocks trained for 200 epochs | |
3) testing-monai-replicate-noADN : contains the monai unet model created Conv2d blocks | |
4) testing-monai-2 : contains the imported monai unet model | |
5) testing-pytorch -best model val : This is a model created using pytorch modules and gave the best dice score and this | |
best performing model is applied on the testing dataset for visualization | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |
miscellaneous | This folder contains the ipymb file(Preprocessing.ipymb) used to preprocess the data and tar_extractor.ipymb used to |
extract the data. | |
------------------- | --------------------------------------------------------------------------------------------------------------------------- |