def train_traffic_net():
download_traffic_net()
trainer = ModelTraining()
trainer.setModelTypeAsResNet()
trainer.setDataDirectory("trafficnet_dataset_v1")
trainer.trainModel(num_objects=4, num_experiments=200, batch_size=32, save_full_model=True, enhance_data=True)
def main():
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory("data/images")
model_trainer.trainModel(num_objects=2,
num_experiments=20,
enhance_data=True,
batch_size=16,
show_network_summary=True)
def main(path_data):
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory(path_data)
model_trainer.trainModel(num_objects=10,
num_experiments=20,
enhance_data=True,
batch_size=32,
show_network_summary=True)
def ImageREcognition():
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory("idenprof")
model_trainer.trainModel(num_objects=10,
num_experiments=100,
enhance_data=True,
batch_size=32,
show_network_summary=True)
def training(image_directory):
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory(training_image_directory)
model_trainer.trainModel(num_objects=1,
num_experiments=200,
enhance_data=True,
batch_size=5,
show_network_summary=True)
def train_func(n):
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory('Dataset')
model_trainer.trainModel(num_objects=n,
num_experiments=100,
enhance_data=True,
batch_size=32,
show_network_summary=True)
def modelIrain(dataDir='data', classNum=2, epochs=100, batch_size=32):
'''
模型训练部分
'''
#创建了ModelTraining类的新实例
model_trainer = ModelTraining()
#将模型类型设置为ResNet
model_trainer.setModelTypeAsResNet()
#设置我们想要训练的数据集的路径
model_trainer.setDataDirectory(dataDir)
#模型训练
'''
num_objects:该参数用于指定图像数据集中对象的数量
num_experiments:该参数用于指定将对图像训练的次数，也称为epochs
enhance_data(可选):该参数用于指定是否生成训练图像的副本以获得更好的性能
batch_size:该参数用于指定批次数量。由于内存限制，需要分批训练，直到所有批次训练集都完成为止。
show_network_summary:该参数用于指定是否在控制台中显示训练的过程
'''
model_trainer.trainModel(num_objects=classNum,
num_experiments=epochs,
enhance_data=True,
batch_size=batch_size,
show_network_summary=True)
print('Model Train Finished!!!')
def modelPredict(model_path='data/models/model_ex-001_acc-0.500000.h5',
class_path='data/json/model_class.json',
pic_path='a.jpg',
classNum=2,
resNum=5):
'''
模型预测部分
prediction_speed[模型加载的速度]:fast faster fastest
'''
prediction = CustomImagePrediction()
prediction.setModelTypeAsResNet()
prediction.setModelPath(model_path)
prediction.setJsonPath(class_path)
prediction.loadModel(num_objects=classNum, prediction_speed='fastest')
prediction, probabilities = prediction.predictImage(
pic_path, result_count=resNum)
for eachPrediction, eachProbability in zip(predictions, probabilities):
print(eachPrediction + " : " + str(eachProbability))
if __name__ == '__main__':
#模型训练
modelTrain(dataDir='data', classNum=2, epochs=10, batch_size=8)
#模型识别预测
modelPredict(model_path='data/models/model_ex-001_acc-0.500000.h5',
class_path='data/json/model_class.json',
pic_path='test.jpg',
classNum=2,
resNum=5)
def train_model(foldername,Model_Type,num_objects=2, num_experiments=1, enhance_data=False, batch_size=1, show_network_summary=True):
model_trainer = ModelTraining()
if Model_Type in "ResNet":
model_trainer.setModelTypeAsResNet()
elif Model_Type in "SqueezeNet":
model_trainer.setModelTypeAsSqueezeNet()
elif Model_Type in "InceptionV3":
model_trainer.setModelTypeAsInceptionV3()
elif Model_Type in "DenseNet":
model_trainer.setModelTypeAsDenseNet()
model_trainer.setDataDirectory(foldername)
model_trainer.trainModel(num_objects=num_objects, num_experiments=num_experiments, enhance_data=enhance_data, batch_size=batch_size, show_network_summary=show_network_summary)
def test_inception_v3_training():
trainer = ModelTraining()
trainer.setModelTypeAsInceptionV3()
trainer.setDataDirectory(data_directory=sample_dataset)
trainer.trainModel(num_objects=10,
num_experiments=1,
enhance_data=True,
batch_size=4,
show_network_summary=True)
assert os.path.isdir(sample_dataset_json_folder)
assert os.path.isdir(sample_dataset_models_folder)
assert os.path.isfile(
os.path.join(sample_dataset_json_folder, "model_class.json"))
assert (len(os.listdir(sample_dataset_models_folder)) > 0)
shutil.rmtree(os.path.join(sample_dataset_json_folder))
shutil.rmtree(os.path.join(sample_dataset_models_folder))
def TrainModel(files, Classes, Epochs, BatchSize):
model_trainer = ModelTraining() #create an instance for de model training
model_trainer.setModelTypeAsResNet(
) #set model to ResNet NN (SqueezeNet, ResNet, InceptionV3 and DenseNet)
model_trainer.setDataDirectory(
files) #folder that contains train and test sets
'''
train model function
number_objects : number of clases
num_experiments : number of iterations (epochs)
Enhance_data (Optional) : if true, creates modified copies of the images to maximize accuracy (but more process cost)
batch_size: number of images that the model trainer will study at once
Show_network_summary (Optional) : if true, shows the structure of the model type
'''
model_trainer.trainModel(num_objects=Classes,
num_experiments=Epochs,
enhance_data=True,
batch_size=BatchSize,
show_network_summary=True)
def train(
self,
epochs=100,
enhance_data=True,
batch_size=16,
show_network_summary=True,
):
click.echo("Start training...")
# Instantiate a ModelTraining object that will be used for model training
trainer = ModelTraining()
# Set the model type of the neural network (it must be the same of the
# prediction)
self._set_proper_model_type(self.model_type, trainer)
# Set the path to the data directory
trainer.setDataDirectory(
os.path.join(self.base_path, self.dataset_name))
# Train the model
trainer.trainModel(
num_objects=self.class_number,
num_experiments=epochs,
enhance_data=enhance_data,
batch_size=batch_size,
show_network_summary=show_network_summary,
)
from imageai.Prediction.Custom import ModelTraining
model_trainer = ModelTraining()
model_trainer.setModelTypeAsResNet()
model_trainer.setDataDirectory(r"D:/python/imageAI_model_train/pets")
model_trainer.trainModel(num_objects=2,
num_experiments=100,
enhance_data=True,
batch_size=16,
show_network_summary=True)
from imageai.Prediction.Custom import ModelTraining
model_trainer = ModelTraining(
) # type of training algorithm in this case basic NN
model_trainer.setModelTypeAsResNet(
) # defines the type of model that will be stored, in this case it will be a simple .h5 file
model_trainer.setDataDirectory(
"idenprof") # where the AI will look for data to train it
model_trainer.trainModel(num_objects=2,
num_experiments=200,
batch_size=32,
show_network_summary=True)
# num_objects is the total number of different types of objects, eg: chef, car, cat
# num_experiments is the total number of epochs or the total number of times an "experiment" is run
# batch_size is the total number of images tested in an epoch
from imageai.Prediction.Custom import ModelTraining
model_trainer = ModelTraining(
) ##create an instance of the ModelTraining class
model_trainer.setModelTypeAsResNet(
) ##set your instance property and start the traning process.
##this function sets the model type of the training instance you created to the ResNet
##model, which means the ResNet algorithm will be trained on your dataset
model_trainer.setDataDirectory(
r"C:\Users\Andreas Thoma\Desktop\Mathimata\EPL445\Project\Tortillas")
## accepts a string which must be the path to the folder that contains the test and train subfolder of your image dataset
model_trainer.trainModel(num_objects=2,
num_experiments=5,
enhance_data=True,
batch_size=32,
show_network_summary=True)
##this is the function that starts the training process. Once it starts, it will create a JSON file in
##the dataset/json folder (e.g Tortillas/json) which contains the mapping of the classes of the dataset.
##The JSON file will be used during custom prediction to produce reults