Merge remote-tracking branch 'refs/remotes/origin/master'

README.md

@@ -72,7 +72,7 @@ source env/bin/activate
 
 **Note**: For further information on how to set up a virtual python environment (also on **Windows**) refer to https://docs.python.org/3/library/venv.html .
 
-When successfully installed, the software outputs the line : "Successfully installed NNC-0.2.2"
+When successfully installed, the software outputs the line : "Successfully installed NNC-0.3.0"
 
 ### Importing the main module
 

create_venv_macos.sh

@@ -0,0 +1,8 @@
+#!/bin/bash
+
+python3 -m venv env
+source env/bin/activate
+pip install --upgrade pip
+pip install -r requirements_macos.txt
+pip install -e .
+deactivate

framework/pytorch_model/__init__.py

@@ -482,7 +482,14 @@ def __init__(self,
                  lr=1e-4,
                  ):
 
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        if torch.cuda.is_available():
+            device = "cuda"
+        elif torch.backends.mps.is_available():
+            device = "mps"
+        else:
+            device = "cpu"
+
+        self.device = torch.device(device)
 
         torch.manual_seed(451)
         torch.backends.cudnn.deterministic = True
@@ -491,7 +498,6 @@ def __init__(self,
         self.learning_rate = lr
         self.epochs = epochs
         self.max_batches = max_batches
-
         self.handle = handler
         if test_set:
             self.test_set = test_set
@@ -517,7 +523,6 @@ def test_model(self,
                    verbose=False
                    ):
 
-        torch.set_num_threads(1)
         Model = copy.deepcopy(self.model)
 
         base_model_arch = Model.state_dict()
@@ -556,7 +561,6 @@ def eval_model(self,
                    verbose=False
                    ):
 
-        torch.set_num_threads(1)
 
         Model = copy.deepcopy(self.model)
 
@@ -595,7 +599,7 @@ def tune_model(
             ft_flag=False,
             verbose=False,     
     ):
-        torch.set_num_threads(1)
+
         verbose = 1 if (verbose & 1) else 0
 
         base_model_arch = self.model.state_dict()

framework/tensorflow_model/__init__.py

@@ -131,6 +131,8 @@ def create_NNC_model_instance_from_file(
         model_struct = loaded_model_struct
 
     if dataset_path and model_struct:
+        if model_name == None and hasattr(model_struct, 'name'):
+            model_name=model_struct.name
         TEFModelExecuter = create_imagenet_model_executer(model_struct=model_struct,
                                                           dataset_path=dataset_path,
                                                           lr=lr,
@@ -161,6 +163,8 @@ def create_NNC_model_instance_from_object(
         model_struct = loaded_model_struct
 
     if dataset_path and model_struct:
+        if model_name == None and hasattr(model_struct, 'name'):
+            model_name=model_struct.name
         TEFModelExecuter = create_imagenet_model_executer(model_struct=model_struct,
                                                           dataset_path=dataset_path,
                                                           lr=lr,
@@ -230,6 +234,10 @@ def __init__(self, model_dict=None):
     def load_model( self, 
                     model_path
                   ):
+
+        try:
+            model_file = tf.keras.models.load_model(model_path)
+        except:
         model_file = h5py.File(model_path, 'r')
 
         try:
@@ -262,26 +270,19 @@ def init_model_from_model_object(   self,
                                         model_object,
                                     ):
         self.model = model_object
-
-        h5_model_path = './temp.h5'
-        model_object.save_weights(h5_model_path)
-        model = h5py.File(h5_model_path, 'r')
-        os.remove(h5_model_path)
 
-        if 'layer_names' in model.attrs:
-            module_names = [n for n in model.attrs['layer_names']]
-
+        weights = model_object.get_weights()
         layer_names = []
-        for mod_name in module_names:
-            layer = model[mod_name]
-            if 'weight_names' in layer.attrs:
-                weight_names = [mod_name+'/'+n for n in layer.attrs['weight_names']]
-                if weight_names:
-                    layer_names += weight_names
+
+        for layer in model_object.layers:
+            mod_name = layer.name
+            if layer.weights != []:
+                for weight in layer.weights:
+                    layer_names.append(mod_name+"/"+weight.name)
 
         model_parameter_dict = {}
-        for name in layer_names:
-            model_parameter_dict[name] = model[name]
+        for i, name in enumerate(layer_names):
+            model_parameter_dict[name] = weights[i]
 
         return self.init_model_from_dict( model_parameter_dict ), model_object
 

framework/use_case_init/__init__.py

@@ -148,47 +148,49 @@ def preprocess(
     ):
         image_size = 224
 
-        if self.__model_name == 'EfficientNetB1':
+        if self.__model_name == 'EfficientNetB1' or self.__model_name == 'efficientnetb1':
             image_size = 240
-        elif self.__model_name == 'EfficientNetB2':
+        elif self.__model_name == 'EfficientNetB2' or self.__model_name == 'efficientnetb2':
             image_size = 260
-        elif self.__model_name == 'EfficientNetB3':
+        elif self.__model_name == 'EfficientNetB3' or self.__model_name == 'efficientnetb3':
             image_size = 300
-        elif self.__model_name == 'EfficientNetB4':
+        elif self.__model_name == 'EfficientNetB4' or self.__model_name == 'efficientnetb4':
             image_size = 380
-        elif self.__model_name == 'EfficientNetB5':
+        elif self.__model_name == 'EfficientNetB5' or self.__model_name == 'efficientnetb5':
             image_size = 456
-        elif self.__model_name == 'EfficientNetB6':
+        elif self.__model_name == 'EfficientNetB6' or self.__model_name == 'efficientnetb6':
             image_size = 528
-        elif self.__model_name == 'EfficientNetB7':
+        elif self.__model_name == 'EfficientNetB7' or self.__model_name == 'efficientnetb7':
             image_size = 600
 
         image, label = self.model_transform(image, label, image_size=image_size)
 
-        if 'DenseNet' in self.__model_name:
+        if 'DenseNet' in self.__model_name or 'densenet' in self.__model_name:
             return tf.keras.applications.densenet.preprocess_input(image), label
-        elif 'EfficientNet' in self.__model_name:
+        elif 'EfficientNet' in self.__model_name or 'efficientnet' in self.__model_name:
             return tf.keras.applications.efficientnet.preprocess_input(image), label
-        elif self.__model_name == 'InceptionResNetV2':
+        elif self.__model_name == 'InceptionResNetV2' or self.__model_name == 'inception_resnet_v2':
             return tf.keras.applications.inception_resnet_v2.preprocess_input(image), label
-        elif self.__model_name == 'InceptionV3':
+        elif self.__model_name == 'InceptionV3' or self.__model_name == "inception_v3":
             return tf.keras.applications.inception_v3.preprocess_input(image), label
-        elif self.__model_name == 'MobileNet':
+        elif self.__model_name == 'MobileNet' or ( 'mobilenet' in self.__model_name and 'v2' not in self.__model_name ):
             return tf.keras.applications.mobilenet.preprocess_input(image), label
-        elif self.__model_name == 'MobileNetV2':
+        elif self.__model_name == 'MobileNetV2' or 'mobilenetv2' in self.__model_name:
             return tf.keras.applications.mobilenet_v2.preprocess_input(image), label
         elif 'NASNet' in self.__model_name:
             return tf.keras.applications.nasnet.preprocess_input(image), label
-        elif 'ResNet' in self.__model_name and 'V2' not in self.__model_name:
+        elif ('ResNet' in self.__model_name and 'V2' not in self.__model_name) or ('resnet' in self.__model_name and 'v2' not in self.__model_name):
             return tf.keras.applications.resnet.preprocess_input(image), label
-        elif 'ResNet' in self.__model_name and 'V2' in self.__model_name:
+        elif ('ResNet' in self.__model_name and 'V2' in self.__model_name) or ('resnet' in self.__model_name and 'v2' in self.__model_name):
             return tf.keras.applications.resnet_v2.preprocess_input(image), label
-        elif self.__model_name == 'VGG16':
+        elif self.__model_name == 'VGG16' or self.__model_name == 'vgg16':
             return tf.keras.applications.vgg16.preprocess_input(image), label
-        elif self.__model_name == 'VGG19':
+        elif self.__model_name == 'VGG19' or self.__model_name == 'vgg19':
             return tf.keras.applications.vgg19.preprocess_input(image), label
-        elif self.__model_name == 'Xception':
+        elif self.__model_name == 'Xception' or self.__model_name == 'xception':
             return tf.keras.applications.xception.preprocess_input(image), label
+        elif 'RegNet' in self.__model_name or 'regnet' in self.__model_name:
+            return tf.keras.applications.regnet.preprocess_input(image), label
 
 
 # supported use cases

requirements_cu11.txt

@@ -4,7 +4,7 @@ scikit-learn>=0.23.1
 tqdm>=4.32.2
 h5py>=3.1.0
 pybind11>=2.6.2
-tensorflow>=2.6.0
+tensorflow[and-cuda]>=2.6.0
 pandas>=1.0.5
 opencv-python>=4.4.0.46
 torch>=1.8.1 

requirements_macos.txt

@@ -0,0 +1,12 @@
+urllib3<2.0
+Click>=7.0
+scikit-learn>=0.23.1
+tqdm>=4.32.2
+h5py>=3.1.0
+pybind11>=2.6.2
+tensorflow>=2.13.0
+tensorflow-metal>=1.0.0
+pandas>=1.0.5
+opencv-python>=4.4.0.46
+torch>=1.12.0
+torchvision>=0.13.1 

setup.py

@@ -49,7 +49,7 @@
 from setuptools.command.build_ext import build_ext
 import setuptools
 
-__version__ = '0.2.2'
+__version__ = '0.3.0'
 
 
 class get_pybind_include(object):