def clip_bboxes_TLBR(bboxes_in, im_width, im_height, warning=True, debug=True):
'''
this function clips bboxes inside the image boundary, the coordinates in the clipped bbox are half-included [x, y)
parameters:
bboxes_in: TLBR format, a list of 4 elements, a listoflist of 4 elements: e.g., [[1,2,3,4], [5,6,7,8]],
a numpy array with shape or (N, 4) or (4, )
im_width/im_height: scalar
outputs:
clipped_bboxes: TLBR format, numpy array with shape of (N, 4)
'''
np_bboxes = safe_bbox(bboxes_in, warning=warning, debug=debug)
if debug:
assert isinteger(im_width) and isinteger(
im_height), 'the image width and height are not correct'
assert bboxcheck_TLBR(np_bboxes, warning=warning,
debug=debug), 'the input bboxes are not good'
clipped_bboxes = np.zeros_like(np_bboxes)
clipped_bboxes[:, 0] = np.maximum(np.minimum(np_bboxes[:, 0], im_width),
0) # x1 >= 0 & x1 <= width, included
clipped_bboxes[:, 1] = np.maximum(np.minimum(np_bboxes[:, 1], im_height),
0) # y1 >= 0 & y1 <= height, included
clipped_bboxes[:, 2] = np.maximum(np.minimum(np_bboxes[:, 2], im_width),
0) # x2 >= 0 & x2 <= width, not included
clipped_bboxes[:,
3] = np.maximum(np.minimum(np_bboxes[:, 3], im_height),
0) # y2 >= 0 & y2 <= height, not included
return clipped_bboxes
def bbox_enlarge(bbox, ratio=0.2, width_ratio=None, height_ratio=None, min_length=128, min_width=None, min_height=None, debug=True):
'''
enlarge the bbox around the edge
parameters:
bbox: N X 4 numpy array, TLBR format
ratio: how much to enlarge, for example, the ratio=0.2, then the width and height will be increased by 0.2 times of original width and height
'''
if debug:
assert bboxcheck_TLBR(bbox), 'the input bounding box should be TLBR format'
if width_ratio is not None and height_ratio is not None:
width = (bbox[:, 2] - bbox[:, 0]) * width_ratio
height = (bbox[:, 3] - bbox[:, 1]) * height_ratio
else:
width = (bbox[:, 2] - bbox[:, 0]) * ratio
height = (bbox[:, 3] - bbox[:, 1]) * ratio
cur_width = bbox[:, 2] - bbox[:, 0]
cur_height = bbox[:, 3] - bbox[:, 1]
if min_height is not None and min_width is not None:
width = max(width, min_width - cur_width)
height = max(height, min_height - cur_height)
else:
width = max(width, min_length - cur_width)
height = max(height, min_length - cur_height)
bbox[:, 0] -= width / 2.0
bbox[:, 2] += width / 2.0
bbox[:, 3] += height / 2.0
bbox[:, 1] -= height / 2.0
return bbox
def bbox2center(bboxes_in, debug=True, vis=False):
'''
convert a bounding box to a point, which is the center of this bounding box
parameter:
bbox: N x 4 numpy array, TLBR format
return:
center: 2 x N numpy array, x and y correspond to first and second row respectively
'''
np_bboxes = safe_bbox(bboxes_in, debug=debug)
if debug:
assert bboxcheck_TLBR(
np_bboxes), 'the input bounding box should be TLBR format'
num_bbox = np_bboxes.shape[0]
center = np.zeros((num_bbox, 2), dtype='float32')
center[:, 0] = (np_bboxes[:, 0] + np_bboxes[:, 2]) / 2.
center[:, 1] = (np_bboxes[:, 1] + np_bboxes[:, 3]) / 2.
# if vis:
# fig = plt.figure()
# plt.scatter(np_bboxes[0, 0], -np_bboxes[0, 1], color='b') # -1 is to convert the coordinate from image to cartesian
# plt.scatter(np_bboxes[0, 2], -np_bboxes[0, 3], color='b')
# center_show = imagecoor2cartesian(center)
# plt.scatter(center_show[0], center_show[1], color='r')
# plt.show()
# plt.close(fig)
return np.transpose(center)
def bbox_TLBR2TLWH(bboxes_in, debug=True): ''' transform the input bounding box with TLBR format to TLWH format parameters: bboxes_in: TLBR format, a list of 4 elements, a listoflist of 4 elements: e.g., [[1,2,3,4], [5,6,7,8]], a numpy array with shape or (N, 4) or (4, ) outputs: bbox_TLWH: N X 4 numpy array, TLWH format ''' np_bboxes = safe_bbox(bboxes_in, debug=debug) if debug: assert bboxcheck_TLBR(np_bboxes, debug=debug), 'the input bounding box should be TLBR format' bbox_TLWH = np.zeros_like(np_bboxes) bbox_TLWH[:, 0] = np_bboxes[:, 0] bbox_TLWH[:, 1] = np_bboxes[:, 1] bbox_TLWH[:, 2] = np_bboxes[:, 2] - np_bboxes[:, 0] bbox_TLWH[:, 3] = np_bboxes[:, 3] - np_bboxes[:, 1] return bbox_TLWH
