Error using segnetLayers with resnet101

Hello everyone,

I have a working neural network when using vgg16. however when I want to change the net I am training to resnet101 (because it is supposed to be faster) I am receiving the following error:

Error using segnetLayers>iCheckDepthOrName (line 470)
Expected name to match one of these values:

'vgg16', 'vgg19'

The input, 'resnet101', did not match any of the valid values.


Error in segnetLayers>iParseInputs (line 390)
depthOrName = iCheckDepthOrName(userInput.depth);

Error in segnetLayers (line 166)
args = iParseInputs(imageSize, numClasses, depth, varargin{:});

Error in dieKIresnet101 (line 119)
lgraph = segnetLayers(imageSize,numClasses,'resnet101'); 

my question is: how to replace segnetLayers with another command or is there anything else I can do? Thank you in advance.

Answer 1

[Srivardhan Gadila]

The argument "model" for the segnetLayers function supports 'vgg16' and 'vgg19' only. You may refer here.

For resnet based segmentation networks you may use deeplabv3plusLayers with the supported networks.

Or you may build a segmentation network by using the pretrained renset101 as encoder. The following may help in the process Transfer Learning, Semantic Segmentation Using Deep Learning.

[Philipp Fischer]

Thank you so Much! using deeplabv3plusLayers helped me and finally it worked.