src/fine_tune.py [341:358]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
):
    # -- init model
    if 'wide_resnet' in model_name:
        encoder = wide_resnet.__dict__[model_name](dropout_rate=0.0)
        hidden_dim = 128
    else:
        encoder = resnet.__dict__[model_name]()
        hidden_dim = 2048
        if 'w2' in model_name:
            hidden_dim *= 2
        elif 'w4' in model_name:
            hidden_dim *= 4

    # -- projection head
    encoder.fc = torch.nn.Sequential(OrderedDict([
        ('fc1', torch.nn.Linear(hidden_dim, hidden_dim)),
        ('bn1', torch.nn.BatchNorm1d(hidden_dim)),
        ('relu1', torch.nn.ReLU(inplace=True)),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



src/paws_train.py [418:434]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
):
    if 'wide_resnet' in model_name:
        encoder = wide_resnet.__dict__[model_name](dropout_rate=0.0)
        hidden_dim = 128
    else:
        encoder = resnet.__dict__[model_name]()
        hidden_dim = 2048
        if 'w2' in model_name:
            hidden_dim *= 2
        elif 'w4' in model_name:
            hidden_dim *= 4

    # -- projection head
    encoder.fc = torch.nn.Sequential(OrderedDict([
        ('fc1', torch.nn.Linear(hidden_dim, hidden_dim)),
        ('bn1', torch.nn.BatchNorm1d(hidden_dim)),
        ('relu1', torch.nn.ReLU(inplace=True)),
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -