Public data and data license

[duducheng]

Dear challenge organizers,

I am interested in this challenge, which addresses an important aspect in medical image anlaysis.

I have 2 questions:

1. Do you forbid any additional data, including ImageNet, COCO or any pretrained 2D / 3D CNNs?
2. What is the data license of the provided MRI and CT scans? Is it allowed to develop other datasets based on yours?

Thank you!

[dzimmerer]

Hi,

thanks for your interest. Most Information can be found on the Website.

For your questions:

1. Sadly due to fairness and cheating, we do not allow training your model on any external datasets (feel free to use the provided data to pretrain your nets).

2. You can find the exact license on the download page. But in short (and completely tentative) as long as you don't try to identify any people in the dataset and keep it under the same license, you are allowed to develop other datasets. However, if you want to use it for training your model for the challenge, remember to not use any external data in your new dataset.

I hope that makes things clearer.
Cheers,

[duducheng]

Hi,

Thanks for your response!

I understand your concern on public data, but I still suggest to have some exceptions, for instance, OK to use torchvision / TF pretrained models, because it is reported that pretraining could improve robustness in uncertainty quantification and adversarial setting, which could also generalize to OOD in 3D medical images.