[Solved] Striding a 4d tensor Non-Continuous?

1

I’m trying to index a 4d tensor as such:
x[batch, :, begin_x:end_x, begin_y:end_y]
(begin_x, begin_y, end_x, end_y are all single integers)
However the resultant image looks almost strange [full img on right, slice on the left]

2017-07-06-014439_919x464_scrot.png919×464 403 KB

When I do the following for a 2d image it works as expected. i.e.:
x[batch, begin_x:end_x, begin_y:end:y]

I tried .contiguous to no avail. Am I just doing something dumb?

2

Issue was that the original image was being reshaped in numpy instead of using a tranpose for the dimensions. Then placing it and extracting from pytorch causing the weird re-org. Moral of the story: always transpose dimensions instead of reshape!