The
aiNeuralNetworkModel type is used to define all the advanced characteristics of a neural network used by
AIDetectModel. The characteristics of this neural network can be defined and changed using different WLanguage properties.
Remark: For more details on the declaration of this type of variable and the use of WLanguage properties, see
Declaring a variable.
// Image declaration
MyImage is Image
MyImage = IMG_Test
// Declare a model
MyAINNModel is aiNeuralNetworkModel
MyAINNModel.Configuration = "MyModel.cfg"
MyAINNModel.TrainedWeights = "MyModel.weights.pb"
MyAINNModel.PixelScaleFactor = 1.0
// Caution: the X and Y dimensions depend on the model.
// If the specified dimensions do not match the model,
// AIDetectModel returns an error.
MyAINNModel.XDimension = 300
MyAINNModel.YDimension = 300
MyAINNModel.AverageIntensityR = 104
MyAINNModel.AverageIntensityG = 117
MyAINNModel.AverageIntensityB = 113
MyAINNModel.RGBColor = True
myMatrixArray is array of 1 array of 1 by 1 by 200 by 7 real
// Run model
myMatrixArray = AIDetectModel(MyAINNModel, MyImage)
Remarks
Properties specific to aiNeuralNetworkModel variables
The following properties can be used to handle a neural network model:
| | |
| Property name | Type used | Effect |
|---|
| AverageIntensityB | Integer | Average intensity of the Blue (B) color in the training data. This property is optional. |
| AverageIntensityG | Integer | Average intensity of the Green (G) color in the training data. This property is optional. |
| AverageIntensityR | Integer | Average intensity of the Red (R) color in the training data. This property is optional. |
| Configuration | Character string | Full path of the file that contains the configuration of the model. Remark: The possible types of configurations are: - Caffe: *.prototxt
- Tensorflow: *.pbtxt
- Darknet: *.cfg
|
| OutputLayerName | Character string | Name of the output neural network layer. |
| PixelScaleFactor | Real | Used to scale pixel values. Set to 1.0 by default (no scaling). |
| RGBColor | Boolean | - True if the provided images are encoded in RGB. In this case, the conversion to BGR encoding will be done automatically.
- False (default value) if the provided images are encoded in BGR.
This property is optional. |
| TrainedWeights | Character string | Full path of the file that contains the trained weights of the model. Remark: The possible types of weights are: - Caffe: *.caffemodel
- Tensorflow: *.pb
- Darknet: *.weights
- Open Neural Network Exchange (ONNX): *.onnx
|
| XDimension | Integer | Image size required by the model: X value. By default, this property corresponds to the width of the image. |
| YDimension | Integer | Image size required by the model: Y value. By default, this property corresponds to the height of the image. |