Update the pretrained VGG16 or not?

[o0t1ng0o]

Hi, @lyndonzheng
I have a question about the optimization of pretrained VGG16 when training the Learned SeSim model.
As shown in the following code, the learning rate multiplies zero.
I am wondering whether the pretrained VGG16 will update its parameters during training?

F-LSeSim/models/sc_model.py

Line 109 in e092e62

self.optimizer_F = torch.optim.Adam([{'params': list(filter(lambda p:p.requires_grad, self.netPre.parameters())), 'lr': self.opt.lr*0.0},

Thank you in advance.

[lyndonzheng]

Hi @o0t1ng0o, we did not update weights of pre-trained VGG16. Indeed, we only select suitable features using the contrastive learning from the pre-trained VGG model. We tried to update all the weights using the contrastive learning, while the performance is unstable under current design.

[o0t1ng0o]

Thank you for your prompt reply.