Problem with monai.data.utils.iter_patch

[Mo-Junyang]

Firstly, I want to crop some overlap 3D patches with sliding window from a medical Image. Then I found the function iter_patch and I checked if it work in a simple way.

    a = np.array([[1,2,3,4],[5,6,7,8],[8,7,6,5],[4,3,2,1]])
    print(a)
    c= iter_patch(a,patch_size=[2,2],start_pos=0,overlap=0.5,copy_back=False)
    for i,b in c:
        print(i,"\n")

but get a wired output like this

[[1 2 3 4]
 [5 6 7 8]
 [8 7 6 5]
 [4 3 2 1]]
[[8 7]
 [4 3]] 

[[7 6]
 [3 2]] 

[[6 5]
 [2 1]] 

[[5 8]
 [1 4]] 

[[4 3]] 

[[3 2]] 

[[2 1]] 

[[1 4]] 

[] 

[] 

[] 

[] 

[] 

[] 

[] 

[] 

but when I change the Input into string, it seems to work well

    a = np.array([["a", "a", "b", "b"], ["a", "a", "b", "b"], ["c", "c", "d", "d"], ["c", "c", "d", "d"]])
    print(a)
    c= iter_patch(a,patch_size=[2,2],start_pos=0,overlap=0.5,copy_back=False)
    for i,b in c:
        print(i,"\n")

[['a' 'a' 'b' 'b']
 ['a' 'a' 'b' 'b']
 ['c' 'c' 'd' 'd']
 ['c' 'c' 'd' 'd']]
[['a' 'a']
 ['a' 'a']] 

[['a' 'b']
 ['a' 'b']] 

[['b' 'b']
 ['b' 'b']] 

[['b' 'a']
 ['b' 'a']] 

[['a' 'a']
 ['c' 'c']] 

[['a' 'b']
 ['c' 'd']] 

[['b' 'b']
 ['d' 'd']] 

[['b' 'a']
 ['d' 'c']] 

[['c' 'c']
 ['c' 'c']] 

[['c' 'd']
 ['c' 'd']] 

[['d' 'd']
 ['d' 'd']] 

[['d' 'c']
 ['d' 'c']] 

[['c' 'c']
 ['a' 'a']] 

[['c' 'd']
 ['a' 'b']] 

[['d' 'd']
 ['b' 'b']] 

[['d' 'c']
 ['b' 'a']] 

So, I want to know what happened, and how can I create some overlap patches and save it in files with monai? Thanks a lot

[KumoLiu]

Hi @Mo-Junyang, you need to ensure the data is channel-first, try something like this:

a = np.array([[[1,2,3,4],[5,6,7,8],[8,7,6,5],[4,3,2,1]]])
print(a)
c= iter_patch(a,patch_size=[None,2,2],start_pos=(0,0),overlap=0,copy_back=False)
for i,b in c:
    print(i,"\n")

You can also try to use PatchDataset and GridPatchDataset

Hope it can help you, thanks!

[Mo-Junyang]

Thank you! That helps a lot! BTW, do you know why it didn't go wrong when the input was str type array, even though I didn't ensure the "channel-first"

[KumoLiu]

Sure, because of the different dtype, a different backend is used.

MONAI/monai/transforms/croppad/functional.py

Line 99 in a1c4371

if mode in {"constant", "reflect", "edge", "replicate", "wrap", "circular"} and img.dtype not in {

Thanks!

[Mo-Junyang]

Thanks a lot, it makes me clearer！