Segment image with different slice length

[talhaanwarch]

Greetings. Is there any way to segment or classify data having images of different slice length (z-axis).
I have the following images' shape
(224,224,110),(224,224,113),(224,224,100),(224,224,107).
Is there any way, without croping or padding it to fix length?

The reason i am asking as we cant crop/pad the test set, as it not available in most competition.

[Nic-Ma]

Hi @talhaanwarch ,

Thanks for your interest here.
I think you can run the inference with batch_size=1 to solve your problem.

Thanks.

[talhaanwarch]

But I want to do training. Is it still hold true for training, I mean can I train volumetric image of different slice length

…

On Sat, May 29, 2021, 7:35 PM Nic Ma ***@***.***> wrote: Hi @talhaanwarch <https://github.com/talhaanwarch> , Thanks for your interest here. I think you can run the inference with batch_size=1 to solve your problem. Thanks. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2218 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AI5FYO5YSOBYZ4OQMMWG4G3TQD3SFANCNFSM45FHC5AQ> .

[Nic-Ma]

If you want to batch your images in PyTorch, the images must have same size. So in order to support images with different size, you can use batch_size=1, it would slower but can support your case.
I would suggest you to pad / crop / resize to unify the image size during training.

Thanks.

[talhaanwarch]

Thanks @Nic-Ma , unifying is the best option but it will fix the input size and I can't unifying the test data, because of it wouldnot match the mask then

[rijobro]

For the test data, you can either crop the mask with the same parameters as the input image, or you can continue using batch_size=1, and then you won't have to crop.