3 Copyright (C) 2018-2019 Intel Corporation
5 Licensed under the Apache License, Version 2.0 (the "License");
6 you may not use this file except in compliance with the License.
7 You may obtain a copy of the License at
9 http://www.apache.org/licenses/LICENSE-2.0
11 Unless required by applicable law or agreed to in writing, software
12 distributed under the License is distributed on an "AS IS" BASIS,
13 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 See the License for the specific language governing permissions and
15 limitations under the License.
17 from __future__ import print_function
20 from argparse import ArgumentParser, SUPPRESS
25 from openvino.inference_engine import IENetwork, IEPlugin
28 def build_argparser():
29 parser = ArgumentParser(add_help=False)
30 args = parser.add_argument_group('Options')
31 args.add_argument('-h', '--help', action='help', default=SUPPRESS, help='Show this help message and exit.')
32 args.add_argument("-m", "--model", help="Required. Path to an .xml file with a trained model.",
33 required=True, type=str)
34 args.add_argument("-i", "--input", help="Required. Path to a folder with images or path to an image files",
35 required=True, type=str, nargs="+")
36 args.add_argument("-l", "--cpu_extension",
37 help="Optional. Required for CPU custom layers. Absolute path to a shared library with the"
38 " kernels implementations.", type=str, default=None)
39 args.add_argument("-pp", "--plugin_dir", help="Optional. Path to a plugin folder", type=str, default=None)
40 args.add_argument("-d", "--device",
41 help="Optional. Specify the target device to infer on; CPU, GPU, FPGA, HDDL or MYRIAD is "
42 "acceptable. The sample will look for a suitable plugin for device specified. Default value is CPU",
43 default="CPU", type=str)
44 args.add_argument("--labels", help="Optional. Labels mapping file", default=None, type=str)
45 args.add_argument("-nt", "--number_top", help="Optional. Number of top results", default=10, type=int)
46 args.add_argument("-ni", "--number_iter", help="Optional. Number of inference iterations", default=1, type=int)
47 args.add_argument("-pc", "--perf_counts", help="Optional. Report performance counters",
48 default=False, action="store_true")
54 log.basicConfig(format="[ %(levelname)s ] %(message)s", level=log.INFO, stream=sys.stdout)
55 args = build_argparser().parse_args()
56 model_xml = args.model
57 model_bin = os.path.splitext(model_xml)[0] + ".bin"
59 # Plugin initialization for specified device and load extensions library if specified
60 plugin = IEPlugin(device=args.device, plugin_dirs=args.plugin_dir)
61 if args.cpu_extension and 'CPU' in args.device:
62 plugin.add_cpu_extension(args.cpu_extension)
64 log.info("Loading network files:\n\t{}\n\t{}".format(model_xml, model_bin))
65 net = IENetwork(model=model_xml, weights=model_bin)
67 if plugin.device == "CPU":
68 supported_layers = plugin.get_supported_layers(net)
69 not_supported_layers = [l for l in net.layers.keys() if l not in supported_layers]
70 if len(not_supported_layers) != 0:
71 log.error("Following layers are not supported by the plugin for specified device {}:\n {}".
72 format(plugin.device, ', '.join(not_supported_layers)))
73 log.error("Please try to specify cpu extensions library path in sample's command line parameters using -l "
74 "or --cpu_extension command line argument")
76 assert len(net.inputs.keys()) == 1, "Sample supports only single input topologies"
77 assert len(net.outputs) == 1, "Sample supports only single output topologies"
79 log.info("Preparing input blobs")
80 input_blob = next(iter(net.inputs))
81 out_blob = next(iter(net.outputs))
82 net.batch_size = len(args.input)
84 # Read and pre-process input images
85 n, c, h, w = net.inputs[input_blob].shape
86 images = np.ndarray(shape=(n, c, h, w))
88 image = cv2.imread(args.input[i])
89 if image.shape[:-1] != (h, w):
90 log.warning("Image {} is resized from {} to {}".format(args.input[i], image.shape[:-1], (h, w)))
91 image = cv2.resize(image, (w, h))
92 image = image.transpose((2, 0, 1)) # Change data layout from HWC to CHW
94 log.info("Batch size is {}".format(n))
96 # Loading model to the plugin
97 log.info("Loading model to the plugin")
98 exec_net = plugin.load(network=net)
100 # Start sync inference
101 log.info("Starting inference ({} iterations)".format(args.number_iter))
103 for i in range(args.number_iter):
105 infer_request_handle = exec_net.start_async(request_id=0, inputs={input_blob: images})
106 infer_request_handle.wait()
107 infer_time.append((time() - t0) * 1000)
108 log.info("Average running time of one iteration: {} ms".format(np.average(np.asarray(infer_time))))
110 perf_counts = infer_request_handle.get_perf_counts()
111 log.info("Performance counters:")
112 print("{:<70} {:<15} {:<15} {:<15} {:<10}".format('name', 'layer_type', 'exet_type', 'status', 'real_time, us'))
113 for layer, stats in perf_counts.items():
114 print("{:<70} {:<15} {:<15} {:<15} {:<10}".format(layer, stats['layer_type'], stats['exec_type'],
115 stats['status'], stats['real_time']))
116 # Processing output blob
117 log.info("Processing output blob")
118 res = infer_request_handle.outputs[out_blob]
119 log.info("Top {} results: ".format(args.number_top))
121 with open(args.labels, 'r') as f:
122 labels_map = [x.split(sep=' ', maxsplit=1)[-1].strip() for x in f]
125 classid_str = "classid"
126 probability_str = "probability"
127 for i, probs in enumerate(res):
128 probs = np.squeeze(probs)
129 top_ind = np.argsort(probs)[-args.number_top:][::-1]
130 print("Image {}\n".format(args.input[i]))
131 print(classid_str, probability_str)
132 print("{} {}".format('-' * len(classid_str), '-' * len(probability_str)))
134 det_label = labels_map[id] if labels_map else "{}".format(id)
135 label_length = len(det_label)
136 space_num_before = (7 - label_length) // 2
137 space_num_after = 7 - (space_num_before + label_length) + 2
138 space_num_before_prob = (11 - len(str(probs[id]))) // 2
139 print("{}{}{}{}{:.7f}".format(' ' * space_num_before, det_label,
140 ' ' * space_num_after, ' ' * space_num_before_prob,
145 if __name__ == '__main__':
146 sys.exit(main() or 0)