def pts_euclidean(input_pts1, input_pts2, warning=True, debug=True):
'''
calculate the euclidean distance between two sets of points
parameters:
input_pts1: 2 x N or (2, ) numpy array, a list of 2 elements, a listoflist of 2 elements: (x, y)
input_pts2: same as above
outputs:
ave_euclidean: averaged euclidean distance
eculidean_list: a list of the euclidean distance for all data points
'''
pts1 = safe_2dptsarray(input_pts1, warning=warning, debug=debug)
pts2 = safe_2dptsarray(input_pts2, warning=warning, debug=debug)
if debug:
assert pts1.shape == pts2.shape, 'the shape of two points is not equal'
assert is2dptsarray(pts1) and is2dptsarray(pts2), 'the input points are not correct'
# calculate the distance
eculidean_list = np.zeros((pts1.shape[1], ), dtype='float32')
num_pts = pts1.shape[1]
ave_euclidean = 0
for pts_index in xrange(num_pts):
pts1_tmp = pts1[:, pts_index]
pts2_tmp = pts2[:, pts_index]
n = float(pts_index + 1)
distance_tmp = math.sqrt((pts1_tmp[0] - pts2_tmp[0])**2 + (pts1_tmp[1] - pts2_tmp[1])**2) # TODO check the start
ave_euclidean = (n - 1) / n * ave_euclidean + distance_tmp / n
eculidean_list[pts_index] = distance_tmp
return ave_euclidean, eculidean_list.tolist()
def pts2bbox(pts, debug=True, vis=False):
'''
convert a set of 2d points to a bounding box
parameter:
pts: 2 x N numpy array, N should >= 2
return:
bbox: 1 x 4 numpy array, TLBR format
'''
if debug:
assert is2dptsarray(pts) or is2dptsarray_occlusion(
pts
), 'the input points should have shape: 2 or 3 x num_pts vs %d x %s' % (
pts.shape[0], pts.shape[1])
assert pts.shape[
1] >= 2, 'number of points should be larger or equal than 2'
bbox = np.zeros((1, 4), dtype='float32')
bbox[0, 0] = np.min(pts[0, :]) # x coordinate of left top point
bbox[0, 1] = np.min(pts[1, :]) # y coordinate of left top point
bbox[0, 2] = np.max(pts[0, :]) # x coordinate of bottom right point
bbox[0, 3] = np.max(pts[1, :]) # y coordinate of bottom right point
# if vis:
# fig = plt.figure()
# pts = imagecoor2cartesian(pts)
# plt.scatter(pts[0, :], pts[1, :], color='r')
# plt.scatter(bbox[0, 0], -bbox[0, 1], color='b') # -1 is to convert the coordinate from image to cartesian
# plt.scatter(bbox[0, 2], -bbox[0, 3], color='b')
# plt.show()
# plt.close(fig)
return bbox
def anno_writer(pts_array,
pts_savepath,
num_pts=68,
anno_version=1,
debug=True):
'''
write the point array to a .pts file
parameter:
pts_array: 2 or 3 x num_pts numpy array
'''
if debug:
assert is_path_exists_or_creatable(
pts_savepath), 'the save path is not correct'
assert (
is2dptsarray(pts_array) or is2dptsarray_occlusion(pts_array)
or is2dptsarray_confidence(pts_array)
) and pts_array.shape[1] == num_pts, 'the input point is not correct'
with open(pts_savepath, 'w') as file:
file.write('version: %d\n' % anno_version)
file.write('n_points: %d\n' % num_pts)
file.write('{\n')
# main content
for pts_index in xrange(num_pts):
if is2dptsarray(pts_array):
file.write('%.3f %.3f %f\n' %
(pts_array[0, pts_index], pts_array[1, pts_index],
1.0)) # all visible
else:
file.write('%.3f %.3f %f\n' %
(pts_array[0, pts_index], pts_array[1, pts_index],
pts_array[2, pts_index]))
file.write('}')
file.close()
def safe_2dptsarray(input_pts, homogeneous=False, warning=True, debug=True):
'''
make sure to copy the pts array without modifying it and make the dimension to 2(3 if homogenous) x N
parameters:
input_pts: a list of 2(3 if homogenous) elements, a listoflist of 2 elements:
e.g., [[1,2], [5,6]], a numpy array with shape or (2, N) or (2, )
homogeneous: the input points are in the homogenous coordinate
outputs:
np_pts: 2 (3 if homogenous) X N numpy array
'''
if homogeneous: dimension = 3
else: dimension = 2
if islist(input_pts):
if islistoflist(input_pts):
if debug:
assert all(
len(list_tmp) == dimension for list_tmp in input_pts
), 'all sub-lists should have length of %d' % dimension
np_pts = np.array(input_pts).transpose()
else:
if debug:
assert len(
input_pts
) == dimension, 'the input pts list does not have a good shape'
np_pts = np.array(input_pts).reshape((dimension, 1))
elif isnparray(input_pts):
input_pts = input_pts.copy()
if input_pts.shape == (dimension, ):
np_pts = input_pts.reshape((dimension, 1))
else:
np_pts = input_pts
else:
assert False, 'only list and numpy array for pts are supported'
if debug:
if homogeneous:
assert is2dptsarray_homogeneous(
np_pts), 'the input pts array does not have a good shape'
else:
assert is2dptsarray(
np_pts), 'the input pts array does not have a good shape'
return np_pts
def pts_rotate2D(pts_array, rotation_angle, im_height, im_width, warning=True, debug=True):
'''
rotate the point array in 2D plane counter-clockwise
parameters:
pts_array: 2 x num_pts
rotation_angle: e.g. 90
return
pts_array: 2 x num_pts
'''
if debug: assert is2dptsarray(pts_array), 'the input point array does not have a good shape'
rotation_angle = safe_angle(rotation_angle, warning=warning, debug=True) # ensure to be in [-180, 180]
if rotation_angle > 0: cols2rotated, rows2rotated = im_width, im_width
else: cols2rotated, rows2rotated = im_height, im_height
rotation_matrix = cv2.getRotationMatrix2D((cols2rotated/2, rows2rotated/2), rotation_angle, 1) # 2 x 3
num_pts = pts_array.shape[1]
pts_rotate = np.ones((3, num_pts), dtype='float32') # 3 x num_pts
pts_rotate[0:2, :] = pts_array
return np.dot(rotation_matrix, pts_rotate) # 2 x num_pts
def pts_conversion_back_bbox(pts_array, bboxes_in, debug=True):
'''
convert pts in the cropped image to the pts in the original image
parameters:
bboxes_in: 1 X 4 numpy array, TLBR or TLWH format
pts_array: 2(3) x N numpy array, N should >= 1
'''
np_bboxes = safe_bbox(bboxes_in, debug=debug)
if debug:
assert is2dptsarray(pts_array) or is2dptsarray_occlusion(
pts_array
) or is2dptsarray_confidence(
pts_array
), 'the input points should have shape: 2 or 3 x num_pts vs %d x %s' % (
pts_array.shape[0], pts_array.shape[1])
assert isbbox(np_bboxes), 'the input bounding box is not correct'
pts_out = pts_array.copy()
pts_out[0, :] = pts_array[0, :] + np_bboxes[0, 0]
pts_out[1, :] = pts_array[1, :] + np_bboxes[0, 1]
return pts_out