Why there is no activation on the last layer of ResNET block?

[IIIBlueberry]

Current implementation of this git ResNET block is [x>Conv>Norm>ReLU>Conv>Norm>Add x] shouldn't there should be an activation layer after 'Add x' so that the next resnet block first Conv input is Activated(relu(x))

Or this resnet block implementation supposed to have pre-activation layer?

[junyanz]

We follow the ResNet block design used by Johnson et al's amazing repo https://github.com/jcjohnson/fast-neural-style, which contains two Conv layers per block. The word "resnet block" might be overladed in different papers. We refer you to the original fast style transfer paper for more details.