Need feature maps of ResNet50

I want to get the feature maps of each input image from each layer after an activation function. The underlying model is ResNet50 (pytorch-cifar100/ at master · weiaicunzai/pytorch-cifar100 · GitHub)

In contrast, the VGG Net (pytorch-cifar100/ at master · weiaicunzai/pytorch-cifar100 · GitHub) has a self.features variable where the feature map extraction is easy.

I cannot use this self.features variable, because I also need to extract features within the nn.Sequential block, which is not possible with self.features on my point of view.

After googling, I have two possibilities:

  1. Hooks
  2. Iterating through the graph by using model.children() as here Use Pytorch and Matplotlib to visualize the features of convolutional neural networks - Programmer Sought

Can someone tell me, what is the best way to the get the feature maps of a ResNet50? I need these features for further computations.

I usually use forward hooks as described here, which can store the intermediate activations. You could then pass these activations to further processing.