Single Shot
PUBLISHED
You can use the Tizen.MachineLearning.Inference.SingleShot
class, to load the existing neural network model or your own specific model from the storage. After loading the model, you can invoke it with a single instance of input data. Then, you can get the inference output result.
The main Machine Learning Inference features are:
-
Managing tensor information
Tensor information is the metadata that contains dimensions and types of tensors. You can configure the input and the output Tensor Information such as name, data type, and dimension.
-
Loading a neural network model and configuring a runtime environment
You can load the neural network model from storage and configure a runtime environment.
-
Invoking the neural network model with input data
After setting up the SingleShot instance with its required information, you can invoke the model with the input data and get the inference output result.
-
Fetching the inference result after invoking
You can fetch the inference result after invoking the respective model.
Prerequisites
To enable your application to use the Machine Learning Inference API functionality:
-
To use the methods and properties of the
Tizen.MachineLearning.Inference.SingleShot
class or its related classes such asTizen.MachineLearning.Inference.TensorsData
andTizen.MachineLearning.Inference.TensorsInfo
, include theTizen.MachineLearning.Inference
namespace in your application:using Tizen.MachineLearning.Inference;
-
If the model file you want to use is located in the media storage or the external storage, the application has to request permission by adding the following privileges to the
tizen-manifest.xml
file:<privileges> <!--To access media storage--> <privilege>http://tizen.org/privilege/mediastorage</privilege> <!--To access, read, and write to the external storage--> <privilege>http://tizen.org/privilege/externalstorage</privilege> </privileges>
Managing Tensor Information
In the example mentioned in this page, the MobileNet v1 model for TensorFlow Lite is used. This model is used for image classification. The input data type of the model is specified as bit width of each Tensor and its input dimension is 3 X 224 X 224
. The output data type of the model is the same as the input datatype but the output dimension is 1001 X 1 X 1 X 1
.
To configure the tensor information, you need to create a new instance of the Tizen.MachineLearning.Inference.TensorsInfo
class. Then, you can add the tensor information such as datatype, dimension, and name (optional) as shown in the following code:
/* Input Dimension: 3 * 224 * 224 */ TensorsInfo in_info = new TensorsInfo(); in_info.AddTensorInfo(TensorType.UInt8, new int[4] { 3, 224, 224, 1 }); /* Output Dimension: 1001 for classification */ TensorsData out_info = new TensorsInfo(); out_info.AddTensorInfo(TensorType.UInt8, new int[4] { 1001, 1, 1, 1 });
Loading Neural Network Model and Configuring Runtime Environment
-
Since the model file is located in the resource directory of your own application, you need to get its absolute path:
string ResourcePath = Tizen.Applications.Application.Current.DirectoryInfo.Resource; string model_path = ResourcePath + "models/mobilenet_v1_1.0_224_quant.tflite";
-
You can load the neural network model from storage and configure a runtime environment with the
Tizen.MachineLearning.Inference.SingleShot
class. The first parameter is the absolute path to the neural network model file. The remaining two parameters are the input and the outputTensorsInfo
instances. If there is an invalid parameter,ArgumentException
is raised:/* Create SingleShot instance with model information */ SingleShot single = new SingleShot(model_path, in_info, out_info);
Invoking Neural Network Model using Input Data
To invoke the neural network model, you need to create the Tizen.MachineLearning.Inference.TensorsData
instance to pass the input data of the model. You can add various types of tensor data, which are already specified in the TensorInfo
instance. However, the maximum size of TensorsData
is 16. If the limit is exceeded, then IndexOutOfRangeException
is raised. Input data is passed in a byte array format, byte[]:
/* Input data for test */ byte[] in_buffer = new byte[3 * 224 * 224 * 1]; /* Set the input tensor data */ TensorsData in_data = in_info.GetTensorsData(); in_data.SetTensorData(0, in_buffer);
After preparing the input data, you can invoke the model and get the inference output result. The SingleShot.Invoke()
method gets the input data to be inferred as a parameter and returns the Tizen.MachineLearning.InferenceTensorsData
instance, which contains the inference result:
/* Invoke the model and get the inference result */ TensorsData out_data = single.Invoke(in_data);
Fetching Inference Result
After calling the Invoke()
method of the Tizen.MachineLearning.Inference.SingleShot
class, the Tizen.MachineLearning.Inference.TensorsData
instance is returned as the inference result. The result can have multiple output data. Therefore, you have to fetch each data using the GetTensorData()
method. If the limit is exceeded, then IndexOutOfRangeException
is raised:
/* Get the first Tensor data from the inference result */ byte[] out_buffer = out_data.GetTensorData(0);
The TensorsData
class is used to send the input data to a neural network model. In addition, it provides the Count
property to get the number of tensors:
/* Get the number of Tensor in TensorsData instance */ var count = out_data.Count;