How to import a pretrained deeplabv3+(xception-71 as backbone) model using MATLAB function importNetw​orkFromTen​sorFlow?

In the help document for importNetworkFromTensorFlow function, the argument is a model folder, and the model folder must include saved_model file with the .pb extension and the learned weights in the variables subfolder.

But in the "TensorFlow DeepLab Model Zoo - GitHub", the downloaded folder does not appear to include saved_model.pb and variables subfolder.

Where can I download the required model folder?

Answer 1

[Sivylla Paraskevopoulou]

Hi Chen, you don't need to import the DeepLab v3+ model from TensorFlow. You can use the deeplabv3plusLayers function to create aDeepLab v3+ model for image segmentation in MATLAB. Set the name-value argument network as 'xception' to define the base network.

[Chen Xinbing]

@Sivylla Paraskevopoulou

Thanks very much!

I have studied the source code of function deeplabv3plusLayers using argument 'xception', I find out that the function deeplabv3plusLayers only returns xception-41 network architecture. (Inside deeplabv3plusLayers's source code, it uses function xception() to return a pre-trained xception-41 network. To execute MATLAB code "analyzeNetwork(layerGraph(xception()))" can confirm the fact.) But I want to use xception-65/71 as the backbone of deeplabv3+, so the function deeplabv3plusLayers does not appear to satisfy my purpose.