|
--- |
|
license: mit |
|
task_categories: |
|
- image-to-image |
|
language: |
|
- en |
|
tags: |
|
- medical |
|
--- |
|
|
|
This dataset is based on the BraTS2023 dataset. |
|
It takes 5 middle slices from each nifti volume of the BraTS2023 dataset after normalizing to a value of (-1,1). |
|
All of these images are `.npy` files and one can load them using the `np.load(FILEPATH).astype(np.float32)`. |
|
We provide the training and the test set which contains 6255 and 1095 files respectively. |
|
|
|
It is highly recommend to create a separate validation set from the training dataset for applications. |
|
We use `Pytorch` to do this. We do this by using the following command. |
|
|
|
```python |
|
seed = 97 |
|
train_dataset, val_dataset = torch.utils.data.random_split( |
|
dataset, lengths=(0.9, 0.1), generator=torch.Generator().manual_seed(seed) |
|
) # dataset is the dataset instance. |
|
``` |
|
|
|
This dataset is actually part of a paper which is under peer-review currently. |
|
It is mainly used for multi-domain medical image to image translation. |
|
|
|
We hope this helps the community. |
|
|
