Swin UNETR with 2D image

[YHYeooooong]

Thank you for sharing great work!

I wonder if the Swin UNETR can training and inferencing with 2D image shapes something like (height, width, channel)?

I knew the proposed Swin UNETR trained with 3D images in the paper, but in the model code, it seems to can work on 2D images.

MONAI/monai/networks/nets/swin_unetr.py

Line 588 in acffc2c

elif len(x_shape) == 4:

Thanks!

[tangy5]

HI @YHYeooooong , thanks for the question.
Yes, the implementation of Swin UNETR support 2D images as input, just like UNet or other segmentation networks.
But as you mentioned, we do not have a demo or preliminary benchmark for 2D images. You are very welcome to use for 2D images and see how it work.

Thanks.

[YHYeooooong]

Thanks for fast reply!
I will try ☺