I'm a little bit confusing about the reshape problem of tf.layers.conv3d, the input require shape is (batch, depth, height, width, channels), and now my input data is .nii file(a heart CT scan), which shape is (512, 512, depth), the third dimension represent the heart slices(each slice is 512x512). there are different depth in different patients.
The question is if I try to reshape the data to (depth, 512, 512) in order to fit the conv3d input, the corresponding items of each slice(512x512) will be different.Will this get a influence to training? Or is any method to remain the same items in each slice?
x = np.arange(32).reshape(2,2,8)
print(x[:,:,0])
y = x.reshape(8,2,2)
print(y[0,:,:])
assume that there are 8 slices in heart scan, and there are 2*2 images in each slices.The x[:,:,0] will be ([[0,8],[16,24]]), another one is ([[0,1],[2,3]]). so there are different items in different slices.
I am just wonder whether there exist some method to reshape into(depth, height, width) and keep remain the same items in each slice. or it doesn't matter if i feed this into tensorflow conv3d layer?