Merge pull request #3157 from anzhella-pankratova/read_model_unify

Wovchena · web-flow · commit 27dbca73fe20 · 2022-02-07T14:00:35.000+03:00
Python demos: unify core.read_model()
diff --git a/demos/3d_segmentation_demo/python/3d_segmentation_demo.py b/demos/3d_segmentation_demo/python/3d_segmentation_demo.py
@@ -245,7 +245,6 @@ def read_image(test_data_path, data_name, sizes=(128, 128, 128), is_series=True,
 def main():
     args = parse_arguments()
 
-    # --------------------------------- 1. Load Plugin for inference engine ---------------------------------
     log.info('OpenVINO Inference Engine')
     log.info('\tbuild: {}'.format(get_version()))
     core = Core()
@@ -262,7 +261,6 @@ def main():
         raise AttributeError("Device {} do not support of 3D convolution. "
                              "Please use CPU, GPU or HETERO:*CPU*, HETERO:*GPU*")
 
-    # --------------------- 2. Read IR Generated by ModelOptimizer (.xml and .bin files) ---------------------
     log.info('Reading model {}'.format(args.path_to_model))
     model = core.read_model(args.path_to_model)
 
@@ -279,12 +277,10 @@ def main():
 
     n, c, d, h, w = model.inputs[0].shape
 
-    # ------------------------------------ 3. Loading model to the plugin -------------------------------------
     compiled_model = core.compile_model(model, args.target_device)
     infer_request = compiled_model.create_infer_request()
     log.info('The model {} is loaded to {}'.format(args.path_to_model, args.target_device))
 
-    # --------------------------------------- 4. Preparing input data -----------------------------------------
     start_time = perf_counter()
     if not os.path.exists(args.path_to_input_data):
         raise AttributeError("Path to input data: '{}' does not exist".format(args.path_to_input_data))
@@ -312,9 +308,7 @@ def main():
         original_data = data_crop
         original_size = original_data.shape[-3:]
 
-    # ---------------------------------------------- 5. Do inference --------------------------------------------
     result = infer_request.infer({input_tensor_name: data_crop})
-    # ---------------------------- 6. Processing of the received inference results ------------------------------
     result = next(iter(result.values()))
     batch, channels, out_d, out_h, out_w = result.shape
 
@@ -375,7 +369,6 @@ def main():
     total_latency = (perf_counter() - start_time) * 1e3
     log.info("Metrics report:")
     log.info("\tLatency: {:.1f} ms".format(total_latency))
-    # --------------------------------------------- 7. Save output -----------------------------------------------
     tiff_output_name = os.path.join(args.path_to_output, 'output.tiff')
     Image.new('RGB', (original_data.shape[3], original_data.shape[2])).save(tiff_output_name,
         append_images=list_img, save_all=True)
diff --git a/demos/colorization_demo/python/colorization_demo.py b/demos/colorization_demo/python/colorization_demo.py
@@ -67,7 +67,7 @@ def main(args):
     core = Core()
 
     log.info('Reading model {}'.format(args.model))
-    model = core.read_model(args.model, args.model.with_suffix(".bin"))
+    model = core.read_model(args.model)
 
     input_tensor_name = 'data_l'
     input_shape = model.input(input_tensor_name).shape
diff --git a/demos/face_recognition_demo/python/ie_module.py b/demos/face_recognition_demo/python/ie_module.py
@@ -19,12 +19,12 @@
 
 
 class Module:
-    def __init__(self, core, model, model_type):
+    def __init__(self, core, model_path, model_type):
         self.core = core
         self.model_type = model_type
-        log.info('Reading {} model {}'.format(model_type, model))
-        self.model = core.read_model(model, model.with_suffix('.bin'))
-        self.model_path = model
+        log.info('Reading {} model {}'.format(model_type, model_path))
+        self.model = core.read_model(model_path)
+        self.model_path = model_path
         self.active_requests = 0
         self.clear()
 
diff --git a/demos/formula_recognition_demo/python/utils.py b/demos/formula_recognition_demo/python/utils.py
@@ -120,11 +120,9 @@ def prerocess_crop(crop, tgt_shape, preprocess_type='crop'):
     return preprocess_image(PREPROCESSING[preprocess_type], bin_crop, tgt_shape)
 
 
-def read_net(model_xml, ie, model_type):
-    model_bin = os.path.splitext(model_xml)[0] + ".bin"
-
-    log.info('Reading {} model {}'.format(model_type, model_xml))
-    return ie.read_model(model_xml, model_bin)
+def read_net(model_path, ie, model_type):
+    log.info('Reading {} model {}'.format(model_type, model_path))
+    return ie.read_model(model_path)
 
 
 def change_layout(model_input):
diff --git a/demos/gpt2_text_prediction_demo/python/gpt2_text_prediction_demo.py b/demos/gpt2_text_prediction_demo/python/gpt2_text_prediction_demo.py
@@ -79,9 +79,8 @@ def main():
     ie = Core()
 
     # read IR
-    model_path = args.model
     log.info('Reading model {}'.format(args.model))
-    model = ie.read_model(model_path)
+    model = ie.read_model(args.model)
 
     # check number inputs and outputs
     if len(model.inputs) != 1:
diff --git a/demos/human_pose_estimation_3d_demo/python/modules/inference_engine.py b/demos/human_pose_estimation_3d_demo/python/modules/inference_engine.py
@@ -18,16 +18,16 @@
 
 
 class InferenceEngine:
-    def __init__(self, net_model_xml_path, device, stride):
+    def __init__(self, model_path, device, stride):
         self.device = device
         self.stride = stride
 
         log.info('OpenVINO Inference Engine')
         log.info('\tbuild: {}'.format(get_version()))
         self.core = Core()
 
-        log.info('Reading model {}'.format(net_model_xml_path))
-        self.model = self.core.read_model(net_model_xml_path)
+        log.info('Reading model {}'.format(model_path))
+        self.model = self.core.read_model(model_path)
 
         required_output_keys = {'features', 'heatmaps', 'pafs'}
         for output_tensor_name in required_output_keys:
@@ -40,7 +40,7 @@ def __init__(self, net_model_xml_path, device, stride):
         self.input_tensor_name = self.model.inputs[0].get_any_name()
         compiled_model = self.core.compile_model(self.model, self.device)
         self.infer_request = compiled_model.create_infer_request()
-        log.info('The model {} is loaded to {}'.format(net_model_xml_path, self.device))
+        log.info('The model {} is loaded to {}'.format(model_path, self.device))
 
     def infer(self, img):
         img = img[0:img.shape[0] - (img.shape[0] % self.stride),
diff --git a/demos/image_inpainting_demo/python/inpainting.py b/demos/image_inpainting_demo/python/inpainting.py
@@ -16,7 +16,7 @@
 
 class ImageInpainting:
     def __init__(self, core, model_path, device='CPU'):
-        model = core.read_model(model_path, model_path.with_suffix('.bin'))
+        model = core.read_model(model_path)
 
         if len(model.inputs) != 2:
             raise RuntimeError("The model expects 2 input layers")
diff --git a/demos/image_retrieval_demo/python/image_retrieval_demo/image_retrieval.py b/demos/image_retrieval_demo/python/image_retrieval_demo/image_retrieval.py
@@ -36,9 +36,8 @@ def __init__(self, model_path, device, cpu_extension):
         if cpu_extension and device == 'CPU':
             core.add_extension(cpu_extension, 'CPU')
 
-        path = '.'.join(model_path.split('.')[:-1])
         log.info('Reading model {}'.format(model_path))
-        self.model = core.read_model(path + '.xml', path + '.bin')
+        self.model = core.read_model(model_path)
         self.input_tensor_name = "Placeholder"
         compiled_model = core.compile_model(self.model, device)
         self.infer_request = compiled_model.create_infer_request()
diff --git a/demos/image_translation_demo/python/image_translation_demo/models.py b/demos/image_translation_demo/python/image_translation_demo/models.py
@@ -12,8 +12,8 @@
 """
 
 class CocosnetModel:
-    def __init__(self, core, model_xml, device='CPU'):
-        model = core.read_model(model_xml)
+    def __init__(self, core, model_path, device='CPU'):
+        model = core.read_model(model_path)
         if len(model.inputs) != 3:
             raise RuntimeError("The CocosnetModel expects 3 input layers")
         if len(model.outputs) != 1:
@@ -38,8 +38,8 @@ def infer(self, input_semantics, reference_image, reference_semantics):
 
 
 class SegmentationModel:
-    def __init__(self, core, model_xml, device='CPU'):
-        model = core.read_model(model_xml)
+    def __init__(self, core, model_path, device='CPU'):
+        model = core.read_model(model_path)
         if len(model.inputs) != 1:
             raise RuntimeError("The SegmentationModel expects 1 input layer")
         if len(model.outputs) != 1:
diff --git a/demos/instance_segmentation_demo/python/instance_segmentation_demo.py b/demos/instance_segmentation_demo/python/instance_segmentation_demo.py
@@ -115,7 +115,7 @@ def main():
 
     # Read IR
     log.info('Reading model {}'.format(args.model))
-    model = core.read_model(args.model, args.model.with_suffix('.bin'))
+    model = core.read_model(args.model)
     image_input, image_info_input, (n, c, h, w), model_type, output_names, postprocessor = check_model(model)
     args.no_keep_aspect_ratio = model_type == 'yolact' or args.no_keep_aspect_ratio
 
diff --git a/demos/machine_translation_demo/python/machine_translation_demo.py b/demos/machine_translation_demo/python/machine_translation_demo.py
@@ -30,13 +30,12 @@ class Translator:
     """ Language translation.
 
     Arguments:
-        model_xml (str): path to model's .xml file.
-        model_bin (str): path to model's .bin file.
+        model_path (str): path to model's .xml file.
         tokenizer_src (str): path to src tokenizer.
         tokenizer_tgt (str): path to tgt tokenizer.
     """
-    def __init__(self, model_xml, model_bin, device, tokenizer_src, tokenizer_tgt, output_name):
-        self.engine = TranslationEngine(model_xml, model_bin, device, output_name)
+    def __init__(self, model_path, device, tokenizer_src, tokenizer_tgt, output_name):
+        self.engine = TranslationEngine(model_path, device, output_name)
         self.max_tokens = self.engine.get_max_tokens()
         self.tokenizer_src = Tokenizer(tokenizer_src, self.max_tokens)
         log.debug('Loaded src tokenizer, max tokens: {}'.format(self.max_tokens))
@@ -65,20 +64,19 @@ class TranslationEngine:
     """ OpenVINO engine for machine translation.
 
     Arguments:
-        model_xml (str): path to model's .xml file.
-        model_bin (str): path to model's .bin file.
+        model_path (str): path to model's .xml file.
         output_name (str): name of output blob of model.
     """
-    def __init__(self, model_xml, model_bin, device, output_name):
+    def __init__(self, model_path, device, output_name):
         log.info('OpenVINO Inference Engine')
         log.info('\tbuild: {}'.format(get_version()))
         core = Core()
 
-        log.info('Reading model {}'.format(model_xml))
-        self.model = core.read_model(model_xml, model_bin)
+        log.info('Reading model {}'.format(model_path))
+        self.model = core.read_model(model_path)
         compiled_model = core.compile_model(self.model, args.device)
         self.infer_request = compiled_model.create_infer_request()
-        log.info('The model {} is loaded to {}'.format(model_xml, device))
+        log.info('The model {} is loaded to {}'.format(model_path, device))
         self.input_tensor_name = "tokens"
         self.output_tensor_name = output_name
         self.model.output(self.output_tensor_name) # ensure a tensor with the name exists
@@ -217,8 +215,7 @@ def parse_input(input):
 
 def main(args):
     model = Translator(
-        model_xml=args.model,
-        model_bin=args.model.with_suffix(".bin"),
+        model_path=args.model,
         device=args.device,
         tokenizer_src=args.tokenizer_src,
         tokenizer_tgt=args.tokenizer_tgt,
diff --git a/demos/multi_camera_multi_target_tracking_demo/python/utils/ie_tools.py b/demos/multi_camera_multi_target_tracking_demo/python/utils/ie_tools.py
@@ -57,14 +57,14 @@ def get_input_shape(self):
         """Returns an input shape of the wrapped IE model"""
         return self.model.inputs[0].shape
 
-    def load_model(self, core, model_xml, device, model_type, num_reqs=1, cpu_extension=''):
+    def load_model(self, core, model_path, device, model_type, num_reqs=1, cpu_extension=''):
         """Loads a model in the Inference Engine format"""
         # Plugin initialization for specified device and load extensions library if specified
         if cpu_extension and 'CPU' in device:
             core.add_extension(cpu_extension, 'CPU')
         # Read IR
-        log.info('Reading {} model {}'.format(model_type, model_xml))
-        self.model = core.read_model(model_xml)
+        log.info('Reading {} model {}'.format(model_type, model_path))
+        self.model = core.read_model(model_path)
 
         if len(self.model.inputs) not in self.get_allowed_inputs_len():
             raise RuntimeError("Supports topologies with only {} inputs, but got {}"
@@ -79,4 +79,4 @@ def load_model(self, core, model_xml, device, model_type, num_reqs=1, cpu_extens
         compiled_model = core.compile_model(self.model, device)
         self.infer_queue = AsyncInferQueue(compiled_model, num_reqs)
         self.infer_queue.set_callback(self.completion_callback)
-        log.info('The {} model {} is loaded to {}'.format(model_type, model_xml, device))
+        log.info('The {} model {} is loaded to {}'.format(model_type, model_path, device))
diff --git a/demos/place_recognition_demo/python/place_recognition_demo/place_recognition.py b/demos/place_recognition_demo/python/place_recognition_demo/place_recognition.py
@@ -36,7 +36,7 @@ def __init__(self, model_path, device, cpu_extension):
             core.add_extension(cpu_extension, 'CPU')
 
         log.info('Reading model {}'.format(model_path))
-        self.model = core.read_model(model_path, model_path.with_suffix('.bin'))
+        self.model = core.read_model(model_path)
         self.input_tensor_name = self.model.inputs[0].get_any_name()
         self.input_size = self.model.input(self.input_tensor_name).shape
         self.nchw_layout = self.input_size[1] == 3
diff --git a/demos/single_human_pose_estimation_demo/python/detector.py b/demos/single_human_pose_estimation_demo/python/detector.py
@@ -2,11 +2,11 @@
 
 
 class Detector:
-    def __init__(self, core, path_to_model_xml, label_class, thr=0.3, device='CPU'):
+    def __init__(self, core, model_path, label_class, thr=0.3, device='CPU'):
         self.thr = thr
         self.label_class = label_class
 
-        self.model = core.read_model(path_to_model_xml, path_to_model_xml.with_suffix(".bin"))
+        self.model = core.read_model(model_path)
         if len(self.model.inputs) != 1:
             raise RuntimeError("Detector supports only models with 1 input layer")
         if len(self.model.outputs) != 1:
diff --git a/demos/single_human_pose_estimation_demo/python/estimator.py b/demos/single_human_pose_estimation_demo/python/estimator.py
@@ -82,8 +82,8 @@ def get_trasformation_matrix(center, scale, output_size):
 
 
 class HumanPoseEstimator:
-    def __init__(self, core, path_to_model_xml, device='CPU'):
-        self.model = core.read_model(path_to_model_xml, path_to_model_xml.with_suffix(".bin"))
+    def __init__(self, core, model_path, device='CPU'):
+        self.model = core.read_model(model_path)
         if len(self.model.inputs) != 1:
             raise RuntimeError("HumanPoseEstimator supports only models with 1 input layer")
         if len(self.model.outputs) != 1:
diff --git a/demos/sound_classification_demo/python/sound_classification_demo.py b/demos/sound_classification_demo/python/sound_classification_demo.py
@@ -143,7 +143,7 @@ def main():
         ie.add_extension(args.cpu_extension, 'CPU')
 
     log.info('Reading model {}'.format(args.model))
-    model = ie.read_model(args.model, args.model[:-4] + ".bin")
+    model = ie.read_model(args.model)
 
     if len(model.inputs) != 1:
         log.error("Demo supports only models with 1 input layer")
diff --git a/demos/text_to_speech_demo/python/models/forward_tacotron_ie.py b/demos/text_to_speech_demo/python/models/forward_tacotron_ie.py
@@ -15,7 +15,6 @@
 """
 
 import logging as log
-import os.path as osp
 
 import numpy as np
 
@@ -110,12 +109,9 @@ def gather(a, dim, index):
                                                   [-1 if i == j else 1 for j in range(a.ndim)]) for i in range(a.ndim)]
         return a[tuple(expanded_index)]
 
-    def load_network(self, model_xml):
-        model_bin_name = ".".join(osp.basename(model_xml).split('.')[:-1]) + ".bin"
-        model_bin = osp.join(osp.dirname(model_xml), model_bin_name)
-        log.info('Reading ForwardTacotron model {}'.format(model_xml))
-        model = self.ie.read_model(model=model_xml, weights=model_bin)
-        return model
+    def load_network(self, model_path):
+        log.info('Reading ForwardTacotron model {}'.format(model_path))
+        return self.ie.read_model(model_path)
 
     def create_infer_request(self, model, path):
         compiled_model = self.ie.compile_model(model, device_name=self.device)
diff --git a/demos/text_to_speech_demo/python/models/mel2wave_ie.py b/demos/text_to_speech_demo/python/models/mel2wave_ie.py
@@ -15,7 +15,6 @@
 """
 
 import logging as log
-import os.path as osp
 
 import numpy as np
 
@@ -64,12 +63,9 @@ def __init__(self, model_upsample, model_rnn, ie, target=11000, overlap=550, hop
         self.mel_len = self.upsample_model.input('mels').shape[1] - 2 * self.pad
         self.rnn_width = self.rnn_model.input('h1.1').shape[1]
 
-    def load_network(self, model_xml):
-        model_bin_name = ".".join(osp.basename(model_xml).split('.')[:-1]) + ".bin"
-        model_bin = osp.join(osp.dirname(model_xml), model_bin_name)
-        log.info('Reading WaveRNN model {}'.format(model_xml))
-        model = self.ie.read_model(model=model_xml, weights=model_bin)
-        return model
+    def load_network(self, model_path):
+        log.info('Reading WaveRNN model {}'.format(model_path))
+        return self.ie.read_model(model_path)
 
     def create_infer_requests(self, model, path, batch_sizes=None):
         if batch_sizes is not None:
@@ -221,12 +217,9 @@ def __init__(self, model, ie, device='CPU', default_width=800):
         self.mel_len = self.model.input('mel').shape[2]
         self.widths = [self.mel_len * (i + 1) for i in range(self.scales)]
 
-    def load_network(self, model_xml):
-        model_bin_name = ".".join(osp.basename(model_xml).split('.')[:-1]) + ".bin"
-        model_bin = osp.join(osp.dirname(model_xml), model_bin_name)
-        log.info('Reading MelGAN model {}'.format(model_xml))
-        model = self.ie.read_model(model=model_xml, weights=model_bin)
-        return model
+    def load_network(self, model_path):
+        log.info('Reading MelGAN model {}'.format(model_path))
+        return self.ie.read_model(model_path)
 
     def create_infer_requests(self, model, path, scales=None):
         if scales is not None:
diff --git a/demos/time_series_forecasting_demo/python/time_series_forecasting_demo.py b/demos/time_series_forecasting_demo/python/time_series_forecasting_demo.py
@@ -30,20 +30,20 @@ class ForecastingEngine:
     """ OpenVINO engine for Time Series Forecasting.
 
     Arguments:
-        model_xml (str): path to model's .xml file.
+        model_path (str): path to model's .xml file.
         input_name (str): name of input blob of model.
         output_name (str): name of output blob of model.
     """
-    def __init__(self, model_xml, input_name, output_name, quantiles):
+    def __init__(self, model_path, input_name, output_name, quantiles):
         device = "CPU"
         log.info('OpenVINO Inference Engine')
         log.info('\tbuild: {}'.format(get_version()))
         core = Core()
-        log.info('Reading model {}'.format(model_xml))
-        model = core.read_model(model_xml)
+        log.info('Reading model {}'.format(model_path))
+        model = core.read_model(model_path)
         compiled_model = core.compile_model(model, device)
         self.infer_request = compiled_model.create_infer_request()
-        log.info('The model {} is loaded to {}'.format(model_xml, device))
+        log.info('The model {} is loaded to {}'.format(model_path, device))
         self.input_tensor_name = input_name
         self.output_tensor_name = output_name
         self.quantiles = quantiles
@@ -159,7 +159,7 @@ def build_argparser():
 def main(args):
     quantiles = args.quantiles.split(",")
     model = ForecastingEngine(
-        model_xml=args.model,
+        model_path=args.model,
         input_name=args.input_name,
         output_name=args.output_name,
         quantiles=quantiles
diff --git a/demos/whiteboard_inpainting_demo/python/utils/ie_tools.py b/demos/whiteboard_inpainting_demo/python/utils/ie_tools.py
