def __display_videos(self):
for v_path, l_path in self.data.items():
cv2.destroyAllWindows()
video = cv2.VideoCapture(v_path)
fps = video.get(cv2.CAP_PROP_FPS)
ms = int((1 / fps) * 1000)
v_name = os.path.basename(v_path)
lc = LabelCreator(self.l_type)
frame_no = 0
cv2.namedWindow(v_name)
while True:
grabbed, frame = video.read()
if not grabbed:
break
bboxes = lc.get_boxes_for_frame(l_path, frame_no)
imgu.draw_text('Frame number: ' + str(frame_no), frame)
if len(bboxes) > 0:
bboxes = bbu.convert_center_to_2points(bboxes)
imgu.draw_bbox(bboxes, frame, color=(0, 0, 255), thickness=5)
old_size = frame.shape[:2]
frame = imgu.resize_img(frame, self.frame_h, self.frame_w)
if len(bboxes) > 0:
# bboxes = bbu.convert_center_to_2points(bboxes)
bboxes = bbu.resize_boxes(bboxes, old_size, (self.frame_h, self.frame_w))
imgu.draw_bbox(bboxes, frame, thickness=1)
cv2.imshow(v_name, frame)
k = cv2.waitKey(ms)
if k == 27:
break
if k == ord('p'):
input('Press enter to continue.')
frame_no += 1
def distance(imgs, labels, new_h, new_w):
x_dist = []
y_dist = []
t_dist = []
for img, label in zip(imgs, labels):
bboxes = du.get_boxes(label)
if bboxes is None:
continue
if bboxes.shape[0] < 2:
continue
w, h = Image.open(img).size
h_new, w_new = su.compute_new_size(h, w, new_h, new_w)
bboxes = bbu.resize_boxes(bboxes, (h, w), (h_new, w_new))
distances = compute_distances(bboxes)
for item in distances:
x_dist.append(item[0])
y_dist.append(item[1])
t_dist.append(item[2])
print('Sanity check. Items in x: %d, items in y: %d, items in t: %d' %
(len(x_dist), len(y_dist), len(t_dist)))
print('min x: %d, min y: %d, min total: %d' %
(np.amin(x_dist), np.amin(y_dist), np.amin(t_dist)))
print('max x: %d, max y: %d, max total: %d' %
(np.max(x_dist), np.max(y_dist), np.max(t_dist)))
print('avg x: %.2f, avg y: %.2f, avg total: %.2f' %
(np.average(x_dist), np.average(y_dist), np.average(t_dist)))
return x_dist, y_dist, t_dist
def get_boxes_for_frame(self, labels_path, frame_no, old_shape=None, new_shape=None):
bboxes = []
if self.l_type == 'txt_okutama':
data = self.read_txt(labels_path)
bboxes = self.get_okutama_bbox(frame_no, data)
if self.l_type == 'txt_virat':
pass
if self.l_type == 'xgtf_mdrone':
data = self.read_xml(labels_path)
bboxes = self.get_minidrone_bbox(frame_no, data)
if self.l_type == 'xgtf_UFC1':
data = self.read_xml(labels_path)
bboxes = self.get_ufc1_bboxes(frame_no, data)
if self.l_type == 'xgtf_UFC2':
data = self.read_xml(labels_path)
bboxes = self.get_ufc2_bboxes(frame_no, data)
if (new_shape is not None) and (old_shape is not None):
if bboxes:
bbu.resize_boxes(bboxes, old_shape, new_shape)
return bboxes
def boxes_dimensions(labels, imgs, new_h, new_w):
boxes_w = []
boxes_h = []
for img, label in zip(imgs, labels):
bboxes = du.get_boxes(label)
if bboxes is None:
continue
if bboxes.shape[0] < 2:
continue
w, h = Image.open(img).size
h_new, w_new = su.compute_new_size(h, w, new_h, new_w)
bboxes = bbu.resize_boxes(bboxes, (h, w), (h_new, w_new))
for item in bboxes:
boxes_w.append(item[2])
boxes_h.append(item[3])
return boxes_w, boxes_h
def __display_images(self):
lc = LabelCreator(self.l_type)
for data, label in self.data.items():
img = cv2.imread(data, cv2.IMREAD_COLOR)
old_size = img.shape[:2]
bboxes = lc.get_boxes_for_image(label, os.path.basename(data))
img = imgu.resize_img(img, self.frame_h, self.frame_w)
if len(bboxes) > 0:
bboxes = bbu.convert_center_to_2points(bboxes)
bboxes = bbu.resize_boxes(bboxes, old_size, img.shape[:2])
imgu.draw_bbox(bboxes, img, color=[0, 0, 255], thickness=1)
else:
continue
img = imgu.pad_img(img, self.frame_h, self.frame_w)
# else:
# continue
cv2.imshow('Image', img)
k = cv2.waitKey(1 * 500)
if k == 27:
break
if k == ord('p'):
input('Press enter to continue.')