You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
hi, suppose i want to add a specific augmentation to the dataset while training, where is the right place to do it?
i noticed that you have a function get_training_transforms in nnUNetTrainer.py and i tried to add my own transformation there but it appears that the input array there (data_dict[self.img_key]) is shaped (12,1,602,602). If i want to apply my transformation on the original input image (RGB) before going into feature dimensions where should i do it?
thank you!
Natalie.
The text was updated successfully, but these errors were encountered:
Hey Natalie,
you already found the right place! The transforms are applied in the order that they are added to the tr_transforms list. The shape you are seeing is (b, c, x, y) where b is the batch size and c is the number of channels. Your dataset seems to only have one channel.
Before you proceed I suggest you update to master. We recently replaced the entire data augmentation pipeline. The new pipeline operates on samples, so the shape in your case will be (1, 602, 602)
Best,
Fabian
the thing is that my input image is shaped (1,512,512). When it get to get_training_transforms the image size is already different. in data loader i have noticed the image might have been resampled - am i right? if so, i want to catch it even before. Our goal is to add stiches to the images randomly so it should be applied on the original image 512*512...
hi, suppose i want to add a specific augmentation to the dataset while training, where is the right place to do it?
i noticed that you have a function get_training_transforms in nnUNetTrainer.py and i tried to add my own transformation there but it appears that the input array there (data_dict[self.img_key]) is shaped (12,1,602,602). If i want to apply my transformation on the original input image (RGB) before going into feature dimensions where should i do it?
thank you!
Natalie.
The text was updated successfully, but these errors were encountered: