How can I convert a vector containing entries [[[int int]] …] into a vector containing entries [[int int] …] in python/numpy?

Question

I have data in a numpy vector that looks like this:

 [[[1119   15]]

 [[1125   27]]

 [[1129   43]]

 [[1131   62]]

 [[1131   87]]

 [[1141  234]]

 ...]

These are supposed to be a set of points that I can use to represent a curve, but instead each point [int, int] seems to be encapsulated inside another vector. IE: I have [[1 1]] instead of [1 1].

This data was given to me by an opencv function cv2.approxPolyDP after I fed it a `contour', and I need to work with it. I think the function basically has given me what it thinks is a set of curves, but here each curve only contains one point [int int] which doesn't really make sense. A curve with one point is not a curve, it's a point.

Is there any way to convert [[int int]] to [int int] in this case?

Answer 1

Look at the shape of this array. It probably is (n, 1, 2) .

reshape it to (n,2) . x.reshape(-1,2) is a handy shortcut, saving you the work of determining n . squeeze also gits rid of the singular dimension.

Answer 2

Probably it is not optimal solution but you can do this:

import numpy as np

# example 

a = np.array( [ [[1119, 15]], [[1125, 27]], [[1129, 43]] ] )

# convert

a = np.array( [ x[0] for x in a ] )

print a

[[1119   15]
 [1125   27]
 [1129   43]]

---

EDIT:

import numpy as np

a = np.array( [ [[1119, 15]], [[1125, 27]], [[1129, 143]] ] )

size = len(a)

a = a.reshape([size,2])

print a

    [[1119   15]
     [1125   27]
     [1129   43]]

Answer 3

You can check if the second parameter in approxPolyDP is too big. Notice that the following code doesn't make it small all the time:

epsilon = 0.1*arcLength(contour,True)
polygon = approxPolyDP(contour, epsilon, True)

When the edge of contour gets noisy, the arc length of the contour returned by arcLength can be very very big, which yields a still very big epsilon after being multiplied by 0.1 and thus make approxPolyDP simplify the entire contour into one single point.