how can i change the SegNet architecture to be based on AlexNet not vgg16

Answer 1

[prabhat kumar sharma]

Hi Salma,

I understand that you are using SegNet for semantic segmentation and you want to use AlexNet instead of VGG-16.

To modify a SegNet architecture to be based on AlexNet instead of VGG16 in MATLAB, you will need to replace the encoder part of the SegNet with the layers from AlexNet, while retaining the decoder part that performs the upsampling and pixel classification.

You can follow the below steps :

1. Load Alexnet

alexNet = alexnet;

2. . Modify AlexNet for SegNet Encoder

AlexNet is designed for image classification, so you need to modify it to serve as an encoder for SegNet. This involves removing the fully connected, softmax, and classification layers, as they are not needed for the encoder part.

encoderLayers = alexNet.Layers(1:end-3);

3. Create SegNet Decoder

decoderLayers = [
    % Add your decoder layers here. Each decoder layer typically corresponds
    % to an encoder layer, but performs the opposite operation (e.g., upsampling instead of pooling).
];

4.Combine Encoder and Decoder

layers = [
    encoderLayers
    decoderLayers
    % Add the final layer / Pixel classification layer.
];

5. Now you can create your final SegNet network using the above layers and train your model.

I hope it helps!