How to add regression layer to DAG network like resnet50 for image regression problem?

Hi all,

Can anyone suggest me how to add regression layer to DAG network like Resnet50 at the end, for Iamge regression problem. I could able to understand how to add regression layer to series network architecure like Alexnet. How to do with DAG network architecture like Resnet50?

Answer 1

[Deepika Ahlawat]

Hi,

To remove the classification layer and use regression layer instread you can use the add/remove/replace APIs.

More information can be found here: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layergraph.html?s_tid=srchtitle

The following code should resolve your issue.

net=resnet50;
net.SortedLayers;
lgraph = layerGraph(net);
newlgraph = removeLayers(lgraph,'ClassificationLayer_fc1000');
newlgraph = removeLayers(newlgraph,'fc1000_softmax');
lgraph = addLayers(newlgraph,regressionLayer('Name','out'));
lgraph = connectLayers(lgraph,'fc1000','out');
snet = assembleNetwork(lgraph);

[Dhanaraj Nelapati]

Hi Deepika,

Thank you for your answer, I could able to solve the problem with your insights. Thank you once again.