def predict(self, input_path, output_path):
K.set_learning_phase(0)
image = cv2.imread(input_path)
resized_image = cv2.resize(image, (self.IMAGE_H, self.IMAGE_W))
resized_image = self.normalize_input(resized_image)
input_image = resized_image.reshape((1, self.IMAGE_H, self.IMAGE_W, 3))
dummy_array = np.zeros((1,1,1,1,self.MAX_BOX_PER_IMAGE,4))
netout = self.model.predict([input_image, dummy_array])[0]
boxes = decode_netout(netout, self.OBJ_THRESHOLD, self.NMS_THRESHOLD, self.ANCHORS, len(self.LABELS))
image = draw_boxes(image, boxes, self.LABELS)
print len(boxes), 'Bounding Boxes Found'
print "File Saved to", output_path
cv2.imwrite(output_path, image)
def predict(self, input_paths, output_paths):
assert len(input_paths) == self.SEQUENCE_LENGTH
x = np.zeros((1, self.SEQUENCE_LENGTH, self.IMAGE_H, self.IMAGE_W, 3))
b = np.zeros(
(1, self.SEQUENCE_LENGTH, 1, 1, 1, 1, self.MAX_BOX_PER_IMAGE, 4))
for i, input_path in enumerate(input_paths):
image = cv2.imread(image_path)
resized_image = cv2.resize(image, (self.IMAGE_H, self.IMAGE_W))
resized_image = normalize(resized_image)
x[0, i] = resized_image.reshape((1, self.IMAGE_H, self.IMAGE_W, 3))
b[0, i] = np.zeros((1, 1, 1, 1, self.MAX_BOX_PER_IMAGE, 4))
netouts = self.model.predict([x, b])[0]
for i, netout in enumerate(netouts):
boxes = decode_netout(netout, self.OBJ_THRESHOLD,
self.NMS_THRESHOLD, self.ANCHORS,
len(self.LABELS))
image = draw_boxes(image, boxes, self.LABELS)
print len(boxes), 'Bounding Boxes Found'
print "File Saved to", output_paths[i]
cv2.imwrite(output_paths[i], image)