Changing final activation function from Relu to Softmax in DenseNet CNN

[LaiaHumbertVidan]

Hi, is there a simple way of changing the activation function of the final layers of the DenseNet CNN from Relu to Sigmoid for a binary classification application in Monai?
Many thanks.

[wyli]

I think you can use model.features and add your own classifier for the features

MONAI/monai/networks/nets/densenet.py

Lines 247 to 250 in 7cfc8ea

	def forward(self, x: torch.Tensor) -> torch.Tensor:
	x = self.features(x)
	x = self.class_layers(x)
	return x