Modify ResNet50 to give Multiple Outputs

1

I am dealing with a multi-label classification problem ,the image belongs to one of the
10 classes from two distinct labels i.e desired output is [batch_size,2,10],how can i modify ResNet50 to Get Multiple outputs

2

If I understand your use case correctly you are dealing with two “labels”, each consisting of 10 classes.
Each sample belongs to one particular class of each label.

I think in that particular use case you could use two linear layers, one for each label, and return these two outputs:

class MyModel(nn.Module):
    def __init__(self, num_classes1, num_classes2):
        super(MyModel, self).__init__()
        self.model_resnet = models.resnet18(pretrained=True)
        num_ftrs = self.model_resnet.fc.in_features
        self.model_resnet.fc = nn.Identity()
        self.fc1 = nn.Linear(num_ftrs, num_classes1)
        self.fc2 = nn.Linear(num_ftrs, num_classes2)

    def forward(self, x):
        x = self.model_resnet(x)
        out1 = self.fc1(x)
        out2 = self.fc2(x)
        return out1, out2
2 Likes
3

Thanks so much ,really helped a lot

5

I am getting the following error
AttributeError: module ‘torch.nn’ has no attribute ‘Identity’

6

Probably you are using an older PyTorch version, as the nn.Identity module was introduced in 1.1.0.
Have a look here for install instructions.

7

I used nn.Sequential() it worked fine ,actually i am implementing all this in Kaggle ,the pytorch version must be an older one.Thanks Peter for all the help I was stuck with this for a long time:yum:

1 Like
8

Hey can you tell me which loss function and optimizer did you use and did you freeze the other layers??

9

Sure , I am using SGD with learning_rate of 0.20 .It is working fine ,try this .Hope it helps you .:smile: