The Convolution block in HighResBlock does not utilize acti and norm parameters thus will used default instance norm layer and prelu activation layer. However, it is different from all other Convolution blocks in HighResNet. Is is a mistake? @wyli