StevenBell's profile - activity

Thanks, I was clearly thinking too hard to notice the save() and load() methods right in front of me!

I'm using Python to train an SVM, and I'd like to save the resulting model for use later. However, I get an error:

TypeError: can't pickle SVM objects

Is there another persistence method I should use? Is there something that I could add to the Python SVM module to make it pickle-able?

Minimal working code is below:

import numpy as np
import cv2
import pickle

points = np.array([[1.0, 2.1], [1, -1], [2, 3], [2, 1]], dtype=np.float32)
labels = np.array([0, 1, 0, 1], dtype=np.float32)

model = cv2.SVM(points, labels)
pickle.dump(model, open("save.pkl", 'w'))

What do you mean by "n vectors from the SVM"? Training an SVM should give you a single model.

Found the solution in the find_obj.py sample. The catch is that the C++ code is expecting a "two-channel or three-channel floating-point array, where each element is a 2D/3D vector", which translates in Python/NumPy to a 3-dimensional array.

import cv2
import numpy as np

a = np.array([[1, 2], [4, 5], [7, 8]], dtype='float32')
h = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype='float32')
a = np.array([a])

pointsOut = cv2.perspectiveTransform(a, h)

I am attempting to use the perspectiveTransform() function in Python to warp a set of points:

import cv2
import numpy as np

a = np.array([[1, 2], [4, 5], [7, 8]], dtype='float32')
h = np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]], dtype='float32')
pointsOut = cv2.perspectiveTransform(a, h)

However, all I get is an error:

cv2.error: /build/buildd/opencv-2.3.1/modules/core/src/matmul.cpp:1916:
error: (-215) scn + 1 == m.cols && (depth == CV_32F || depth == CV_64F)
in function perspectiveTransform

I assume I'm passing the arrays in the wrong format somehow, but I've tried quite a few different combinations of array shapes and data types with no success. What's the right way to do this?