search

chaoyi-wu / RadFM

The official code for "Towards Generalist Foundation Model for Radiology by Leveraging Web-scale 2D&3D Medical Data".
315 stars 32 forks source link

What is the shape of the 3D data in your dataset? #7

Open baifanxxx opened 10 months ago

baifanxxx commented 10 months ago

Hi Wu,

Thank you for your great contribution to the community. I have a small question about the data shape. I find this operation here, it seems that the raw 3D CT shape is N,C,H,W,D. In image_dict, “image” shape is C,H,W,D after indexing. However, if contain_nan here, the image shape is C,H,W,D, which can not be indexed on image.shape[0]. I don't know if I'm thinking right here. I just want to figure out what the shape of "image" in image_dict is, and whether their shapes are uniform regardless of whether they are 2D images or 3D images.

Looking forward to your reply. Best regards, BAI Fan

chaoyi-wu commented 10 months ago

Yes，it is a mistake, for nan cases, it should be (1,3,512,512,4) to fill the case so that the progress can be carried out normally. Actually, the operation in if contain_nan will never be done since we have checked and dismissed all the cases contain nan which I think will be a better way.

In image_dict, it contains N images and each image could be (C,H,W,D) for 3D or (C,H,W) for 2D. They can be mixed.