How do I specify crop images in order?

[cvlearn913]

Monai has the following transforms, I have a (512,512,512) size of data that I want to crop to 4 patches in order. The ROI locations is that [0-127,0-127,0-127],[128-255,128-255,128-255],[256-384,256-384,256-384],[385-511,385-511,385-511]. What should I do?

SpatialCrop
CenterSpatialCrop
RandSpatialCrop
RandSpatialCropSamples
CropForeground
RandWeightedCrop
RandCropByPosNegLabel
RandCropByLabelClasses
ResizeWithPadOrCrop
BoundingRect
RandScaleCrop
CenterScaleCrop

[Nic-Ma]

Hi @cvlearn913 ,

Thanks for your interest here.
I think you can (1) add channel dim, (2) copy the original image 3 times, (3) apply SpatialCrop transform to them with expected ROI start / end.

Thanks.

[cvlearn913]

Thank you sincerely for your kind regards！ @Nic-Ma But it seems that dictionary spatialcropd functions do not support multiple clipping. I want to crop and get four samples, but only get one sample.

transforms = Compose(
    [     LoadImaged(keys=["image", "label"]),
          EnsureChannelFirstd(keys=["image", "label"]),
          Resized(keys=["image","label"],spatial_size=[512,512,512]),
          SpatialCropd(keys=["image","label"],roi_size=[128,128,128],roi_slices=[slice(0,128,1),slice(0,128,1),slice(0,128,1)]),
          # The data has already changed to 128*128*128
          EnsureTyped(keys=["image", "label"]),
            
    ]
)

[cvlearn913]

Yeah, I set the roi_start ，roi_end all the same for testing

[Nic-Ma]

Cool!

[AnupKumarGupta]

Hello. Thank you for your solution. I too, have a similar query, but instead of just 3 patches, I require 64 patches. Would there be any other way to do so? Or should I copy the original image 64 times? Thanks in advance.

[KumoLiu]

Hi @AnupKumarGupta, I think you could try to use GridPatch and SplitDim.

MONAI/monai/transforms/spatial/array.py

Line 3176 in 6803061

class GridPatch(Transform):

MONAI/monai/transforms/utility/array.py

Line 334 in 6803061

class SplitDim(Transform):

A simple example here:

from monai.transforms import GridPatch, SplitDim, Compose

trans = Compose([
    GridPatch(num_patches=64),
    SplitDim(dim=0)
])
out = trans(data)

Hope it can help you, thanks!

[AnupKumarGupta]

@KumoLiu Thanks a lot.