def freeze_graph_test(pb_path, outnode):
with tf.Graph().as_default():
output_graph_def = tf.GraphDef()
with open(pb_path, "rb") as f:
output_graph_def.ParseFromString(f.read())
for node in output_graph_def.node:
if node.op == 'Conv2D' and 'explicit_paddings' in node.attr:
del node.attr['explicit_paddings']
if node.op == 'ResizeNearestNeighbor' and 'half_pixel_centers' in node.attr:
del node.attr['half_pixel_centers']
tf.import_graph_def(output_graph_def, name="")
with tf.Session() as sess:
summary_writer = tf.summary.FileWriter('mylog/new2')
summary_writer.add_graph(tf.get_default_graph())
sess.run(tf.global_variables_initializer())
img = cv2.imread('004650.jpg')
originimg = img
orishape = img.shape
img = tools.img_preprocess2(img, None, (INPUTSIZE, INPUTSIZE),
False)[np.newaxis, ...]
img = img.astype(np.float32)
outbox = sess.graph.get_tensor_by_name('YoloV3/output/boxconcat:0')
inputdata = sess.graph.get_tensor_by_name("input/input_data:0")
outbox = sess.run(outbox, feed_dict={inputdata: img})
outbox = np.array(postprocess(outbox, INPUTSIZE, orishape[:2]))
originimg = tools.draw_bbox(originimg, outbox, CLASSES)
cv2.imwrite('004650_detected.jpg', originimg)
def __parse_annotation(self, annotation):
"""
读取annotation中image_path对应的图片，并将该图片进行resize(不改变图片的高宽比)
获取annotation中所有的bbox，并将这些bbox的坐标(xmin, ymin, xmax, ymax)进行纠正，
使得纠正后bbox在resize后的图片中的相对位置与纠正前bbox在resize前的图片中的相对位置相同
:param annotation: 图片地址和bbox的坐标、类别，
如：'image_path xmin,ymin,xmax,ymax,class_ind xmin,ymin,xmax,ymax,class_ind ...'
:return: image和bboxes
bboxes的shape为(N, 5)，其中N表示一站图中有N个bbox，5表示(xmin, ymin, xmax, ymax, class_ind)
"""
line = annotation.split()
image_path = line[0]
image = np.array(cv2.imread(image_path))
bboxes = np.array([map(int, box.split(',')) for box in line[1:]])
image, bboxes = dataAug.random_horizontal_flip(np.copy(image),
np.copy(bboxes))
image, bboxes = dataAug.random_vertical_flip(np.copy(image),
np.copy(bboxes))
image, bboxes = dataAug.random_rotation90_flip(np.copy(image),
np.copy(bboxes))
image, bboxes = dataAug.random_crop(np.copy(image), np.copy(bboxes))
image, bboxes = tools.img_preprocess2(
np.copy(image), np.copy(bboxes),
(self.__train_input_size, self.__train_input_size), True)
return image, bboxes
def __predict(self, image, test_input_size, valid_scale):
org_image = np.copy(image)
org_h, org_w, _ = org_image.shape
cur_milli_time = lambda: int(round(time.time() * 1000))
start_time = cur_milli_time()
yolo_input = tools.img_preprocess2(image, None,
(test_input_size, test_input_size),
False)
yolo_input = yolo_input[np.newaxis, ...]
self.__time_pre += (cur_milli_time() - start_time)
start_time = cur_milli_time()
pred_sbbox, pred_mbbox, pred_lbbox = self.__sess.run(
[self.__pred_sbbox, self.__pred_mbbox, self.__pred_lbbox],
feed_dict={
self.__input_data: yolo_input,
self.__training: False
})
self.__time_inf += (cur_milli_time() - start_time)
start_time = cur_milli_time()
pred_bbox = np.concatenate([
np.reshape(pred_sbbox, (-1, 5 + self._num_classes)),
np.reshape(pred_mbbox, (-1, 5 + self._num_classes)),
np.reshape(pred_lbbox, (-1, 5 + self._num_classes))
],
axis=0)
bboxes = self.__convert_pred(pred_bbox, test_input_size,
(org_h, org_w), valid_scale)
self.__time_pos += (cur_milli_time() - start_time)
return bboxes
def __parse_annotation(self, annotation):
line = annotation.split()
image_path = line[0]
image = np.array(cv2.imread(image_path))
bboxes = np.array([box.split(',') for box in line[1:]], dtype=int)
image, bboxes = dataAug.random_horizontal_flip(np.copy(image), np.copy(bboxes))
image, bboxes = dataAug.random_crop(np.copy(image), np.copy(bboxes))
image, bboxes = dataAug.random_translate(np.copy(image), np.copy(bboxes))
image, bboxes = tools.img_preprocess2(np.copy(image), np.copy(bboxes),
(self.__train_input_size, self.__train_input_size), True)
return image, bboxes
def __predict(self, image, test_input_size, valid_scale):
org_image = np.copy(image)
org_h, org_w, _ = org_image.shape
yolo_input = tools.img_preprocess2(image, None,
(test_input_size, test_input_size),
False)
yolo_input = yolo_input[np.newaxis, ...]
pred_mbbox, pred_lbbox = self.__sess.run(
[self.__pred_mbbox, self.__pred_lbbox],
feed_dict={
self.__input_data: yolo_input,
self.__training: False
})
pred_bbox = np.concatenate([
np.reshape(pred_mbbox, (-1, 5 + self._num_classes)),
np.reshape(pred_lbbox, (-1, 5 + self._num_classes))
],
axis=0)
bboxes = self.__convert_pred(pred_bbox, test_input_size,
(org_h, org_w), valid_scale)
return bboxes