Obtaining opencv UMat shape in python

edit

looking at

>>> help(cv2.UMat)

there clearly is no way to do so, without calling get() (which does a copy to a cpu numpy array)

but: whatever there is in your UMat, it has been a numpy array before, and you probably should have taken the shape from that earlier on.

Well, not exactly. There are some cases when shape of result is unknown. In my case it is result of function: cv2.findNonZero(). I have to repeat this function few hundred times per second and calling UMat.get() every time cost a lot. It is very inconvenient that UMat class is missing such basic method.

well, what about looking at the docs and safely assume:

src.shape == dst.shape

? (but again i do understand your issue here, it's somewhat inconsistent)

Quick test:

x =cv2.UMat(np.array( [1,2,3,0,0,0,1,2,0,0,2,3,0,0,5,0,0,5], dtype=np.uint8))
x.get().shape
(18, 1)
cv2.findNonZero(x).get().shape
(9, 1, 2)

src.shape == dst.shape is not true.

ok. your point ;)

pr 13957