Request for Release the CT Training Data

[xxliang99]

Hi Dr. Chen,
Thank you for your contributions! I would like to reproduce the CT100 result but found the CT training data missing in the dataset folder. The code says:

class CTData(Dataset):
    def __init__(self, mode='train', root_dir='./dataset/CT/CT100_128x128.mat'):
        mat_data = scio.loadmat(root_dir)

But the dataset repository only contains the Urban100 data. I only found 18.3 GB dicom images in the TCGA link given in paper, while I am wondering if there is any metrics in selecting CT slices for training. Therefore, I appreciate it if you would release the CT data you utilized for training.

Thanks for your time!

Best,
Vivian

[edongdongchen]

Hi Dr. Chen, Thank you for your contributions! I would like to reproduce the CT100 result but found the CT training data missing in the dataset folder. The code says:

class CTData(Dataset):
    def __init__(self, mode='train', root_dir='./dataset/CT/CT100_128x128.mat'):
        mat_data = scio.loadmat(root_dir)

But the dataset repository only contains the Urban100 data. I only found 18.3 GB dicom images in the TCGA link given in paper, while I am wondering if there is any metrics in selecting CT slices for training. Therefore, I appreciate it if you would release the CT data you utilized for training.

Thanks for your time!

Best, Vivian

Hi Vivian,

Yes, we just used a tiny subset of TCGA dataset, please download this public subset 'CT100' from here, and the 100 dicom images (458.15 MB) can be found in the folder 'dicom_dir', by then just resize it and save it - you don't really need to download the whole TCGA dataset.

You may want to check the updated ctdb.py and the uploaded .mat file. Let me know if it doesn't help.

Bests,
Dongdong