sohonjit/brats2023_multidomain_i2i

---
license: mit
task_categories:
- image-to-image
language:
- en
tags:
- medical
---

## Dataset Description

- **Paper:** Under Review.
- **Point of Contact:** Arijit Ghosh, [e-mail]

### Dataset Summary
This dataset is based on the BraTS2023 dataset and is supposed to be used for Multi-domain Image-to-Image Translation task. 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 for each domain. These are actually 4 domains, and are named accordingly.

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.

We hope this helps the community.

原始信息汇总

数据集描述

数据集概述

该数据集基于BraTS2023数据集，旨在用于多域图像到图像转换任务。数据集从每个nifti体积中提取5个中间切片，并将其归一化至(-1,1)的值范围。所有图像文件均为.npy格式，可以使用np.load(FILEPATH).astype(np.float32)进行加载。数据集提供了训练集和测试集，分别包含6255和1095个文件，对应4个不同的域。

强烈建议从训练数据集中创建一个单独的验证集。使用Pytorch可以通过以下命令实现：

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是数据集实例。

该数据集是当前正在同行评审的论文的一部分。