How forward() method is used when it have more than one two input parameters in pytorch

user_5400111 0
Viewed 5447

Can someone tell me the concept behind the multiple parameters in forward() method? Generally, the implementation of forward() method has two parameters

  1. self
  2. input

if a forward method has more than these parameters how PyTorch is using the forward method.

Let's consider this codebase: https://github.com/bamps53/kaggle-autonomous-driving2019/blob/master/models/centernet.py here online 236 authors have used forward method with two more parameters:

  1. centers
  2. return_embeddings

I am unable to find a single article that can answer my query on what condition Line 254(return_embeddings:) and Line 257(if centers is not None:) will execute. As per my knowledge forward, the method is internally called by nn module. Can someone please put some lights on this?

tensor pytorch deep-learning neural-network python
1 Answers

Forward function set by you. That means you can add more parameters as you want. For example, you could add inputs as shown below

def forward(self, input1, input2, input3):
    x = self.layer1(input1)
    y = self.layer2(input2)
    z = self.layer3(input3)

    net = torch.cat((x,y,z),1)
     
    return net

You have to control your parameters while feeding the network. Layers couldn't be feed with more than a parameter. Therefore, you need to extract features from your input one by one and concatenate with torch.cat((x,y),1)(1 for dimension) them.

user_86282930
Related