def main(image_file):
detector = Detector(weight_dir=MODEL_WEIGHT_SAVE_DIR,
mode=3,
min_face_size=24)
input_img_full_path = TEST_INPUT_IMG_DIR + '/' + image_file
output_img_full_path = TEST_OUTPUT_IMG_DIR + '/' + image_file
image = cv2.imread(input_img_full_path)
bbox, bboxes, landmarks = detector.predict(image)
print('bboxes-shape---:', bboxes.shape)
print('landmarks-shape---:', landmarks.shape)
for bbox in bboxes:
#print('bbox score--:',bbox[4])
#cv2.putText(image,str(np.round(bbox[4],2)),(int(bbox[0]),int(bbox[1])),cv2.FONT_HERSHEY_TRIPLEX,1,color=(255,0,255))
cv2.rectangle(image, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (0, 0, 255))
for landmark in landmarks:
#print('landmark-shape---:',landmark.shape)
#print('landmark----:',landmark)
for i in range(0, 5):
cv2.circle(image,
(int(landmark[2 * i]), int(int(landmark[2 * i + 1]))),
3, (0, 0, 255))
#break
cv2.imwrite(output_img_full_path, image)
cv2.imshow('yy', image)
cv2.waitKey(0)
def main(input_net_name):
net_name = input_net_name
assert net_name in NET_NAMES
images_dir = WIDER_FACE_IMG_DIR
annotation_file = WIDER_FACE_ANNO_FILE
out_dir = GAN_DATA_ROOT_DIR
out_dir = '{}/{}'.format(out_dir, net_name)
if net_name == 'r_net':
mode = 1
elif net_name == 'o_net':
mode = 2
#images_path images_jpg bboxes
dataset = load_widerface_dataset(images_dir, annotation_file)
detector = Detector(weight_dir= MODEL_WEIGHT_SAVE_DIR, mode=mode)
bboxes_all = []
#p_net 一次测一张图片，注意，其返回可能会有多个，因为图片中可以包含多张面，而且还有图片金字塔
print('data img len --:',len(dataset['images_jpg']))
for img_jpg in dataset['images_jpg']:
_, bboxes, _ = detector.predict(img_jpg)
bboxes_all.append(bboxes)
bboxes_all = np.array(bboxes_all)
print('predict over---')
if not os.path.exists(out_dir):
os.mkdir(out_dir)
detections_path = os.path.join(out_dir, 'middle_wideface_data_det.pkl')
with open(detections_path, 'wb') as f:
pickle.dump({
'bboxes': bboxes_all,
}, f)
save_hard_examples(net_name, out_dir, dataset, detections_path)