PyTorch weight initialization for Alexnet and VGG model

vision
1

I am new to pytorch, and I am confused when I go through the vision models of Alexnet and VGG, for Alexnet, there is no weight initialization function, and other models have., such as for VGG, there is a function def _initialize_weights(self). So my question is, why?

2

Hmm, maybe because it’s too old? But you can generate them by yourself by apply the init rule.

def init_weights(m):
    if isinstance(m, nn.Linear):
        torch.nn.init.xavier_uniform(m.weight)
        m.bias.data.fill_(0.)
    elif isinstance(m, nn.Conv2d):
        torch.nn.init.xavier_uniform(m.weight)

model = torchvisions.models.alexnet()
model.apply(init_weights)
1 Like