Pretrained model as encoder

#1

Hi,
I implemented a autoencoder , and use pretrained model resnet as encoder and the decoder is a series of convTranspose.
Is necessary to apply “init_weights” to autoencoder?
If I use “init_weights” the weights of pretrained model also modified?

Ty.

#2

You can create a separate class for a decoder and init_weights inside this class. Then import it to autoencoder class.

For examples,

class Encoder(nn.Module):
    def __init__(self):
        # load pretrained here

    def forward(self, x):
        # do the thing of encode 

class Decoder(nn.Module):
    def __init__(self):
        # define weigths
        # init weight as random

    def forward(self, x):
        # do the thing to decode


class AutoEncoder(nn.Module):
    def __init__(self):
        self.enc = Encoder()
        self.dec = Decoder()

    def forward(self, x):
        x = self.enc(x)
        return self.dec(x)