Convert a LayerGraph into a DAGNetwork - Validate and Initialize the Net

In theory, the function assembleNetwork() should convert a LayerGraph into a DAGNetwork.

Yet in practice, it doesn't always work:

numRows = 64;
numCols = 64;
numChannels = 3;
numClasses = 3;
assembleNetwork(unetLayers([numRows, numCols, numChannels], numClasses));

Warning: Network issues detected.

Caused by:
Layer 'Segmentation-Layer': Empty Classes property. Classes will be set to categorical(1:N), where N is the number of classes.

Error using assembleNetwork (line 47)
Invalid network.

Caused by:
Layer 'Bridge-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Bridge-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Bridge-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Bridge-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-1-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-1-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-1-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-1-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-1-UpConv': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-1-UpConv': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-2-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-2-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-2-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-2-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-2-UpConv': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-2-UpConv': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-3-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-3-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-3-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-3-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-3-UpConv': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-3-UpConv': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-4-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-4-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-4-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-4-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Decoder-Stage-4-UpConv': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Decoder-Stage-4-UpConv': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-1-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-1-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-1-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-1-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-2-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-2-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-2-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-2-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-3-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-3-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-3-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-3-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-4-Conv-1': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-4-Conv-1': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Encoder-Stage-4-Conv-2': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Encoder-Stage-4-Conv-2': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'Final-ConvolutionLayer': Empty Weights property. Specify a nonempty value for the Weights property.
Layer 'Final-ConvolutionLayer': Empty Bias property. Specify a nonempty value for the Bias property.
Layer 'ImageInputLayer': Empty Mean property. For an image input layer with 'zerocenter' normalization, specify a nonempty value for the Mean property.

My question is, how can I convert myNet = unetLayers([numRows, numCols, numChannels], numClasses) into an initialized and validated network so I can use it for prediction (It will be garbage in garbage out as the net is randomly initialized, but I'm after the action of the conversion)?

Answer 1

[yanqi liu]

yes，sir，may be use assembleNetwork for pretrained layers，if make an init empty unet，may be just use

clc; clear all; close all;
numRows = 64;
numCols = 64;
numChannels = 3;
numClasses = 3;
lgraph = unetLayers([numRows, numCols, numChannels], numClasses)

lgraph = 
  LayerGraph with properties:

         Layers: [58×1 nnet.cnn.layer.Layer]
    Connections: [61×2 table]
     InputNames: {'ImageInputLayer'}
    OutputNames: {'Segmentation-Layer'}

% assembleNetwork(lgraph)
plot(lgraph)

[Royi Avital]

I am after being able to predict with the net without training, not plotting its architecture.