3.5.3. Python代码解析

3.5.3.1. Case 0

在本例中，我们使用的 bmodel 是一个动态模型，它的输入张量尺寸是可变的。 bmodel 中有三个具体模型：PNet、RNet、ONet。 其中，PNet 的输入张量的高和宽可变，RNet 和 ONet 的输入张量的 batch 可变。

# init Engine to load bmodel and allocate input and output tensors
engine = sail.Engine(bmodel_path, 0, sail.SYSIO)
# init preprocessor and postprocessor
preprocessor = PreProcessor([127.5, 127.5, 127.5], 0.0078125)
postprocessor = PostProcessor([0.5, 0.3, 0.7])
# read image
image = cv2.imread(input_path).astype(np.float32)
image = cv2.transpose(image)
# run PNet, the first stage of MTCNN
boxes = run_pnet(engine, preprocessor, postprocessor, image)
if np.array(boxes).shape[0] > 0:
  # run RNet, the second stage of MTCNN
  boxes = run_rnet(preprocessor, postprocessor, boxes, image)
  if np.array(boxes).shape[0] > 0:
    # run ONet, the third stage of MTCNN
    boxes, points = run_onet(preprocessor, postprocessor, boxes, image)
# print detected result
for i, bbox, prob in zip(range(len(boxes)), boxes, probs):
  print("Face {} Box: {}, Score: {}".format(i, bbox, prob))