This component can be used for the encoding layers of a U-Net for 2D data. The following layers are used:
* 2D Convolution Layer
* Dropout Layer
* 2D Convolution Layer
* Max Pooling 2D Layer
An input is required for:
* Filter size for the Convolution layers, e.g. 16
* Activation Function for the Convolution layers, e.g. ELU
* Kernel Size for the Convolution layers, e.g. 3,3
* Dropout Rate for the Dropout layer, e.g. 0.1
* Random seed for the Dropout layer and for the Kernel Initializer in Convolution layers, e.g. 12345
* Pool Size (and Strides) for the Max Pooling layer, e.g. 2,2
The following settings are fixed and passed to the U-Net 2D - Decoding Layer component:
* Strides for the Convolution layers are set to 1,1
* Padding for the Convolution layers is set to "SAME"
* Kernel Initializer for the Convolution layers is set to "He Normal"
The corresponding U-Net 2D - Decoding Layer component can be used to create a complete U-Net.
The required extensions:
- KNIME Deep Learning - Keras Integration
- Type: Keras Deep Learning NetworkKeras ModelEither a Keras Input Layer or a previous U-Net - Encoding Layer