The reviewer further mentions relevant work around deformable convolutions and MSDNet which we will add to19 ourdiscussions. The24 reviewer mentions a parallel between our CondConv approach and Inception modules, but the methods are quite25 different. The reviewer then mentions specific questions with our discussion ofWi. In our CondConv approach, theWi are32 tensors of the same shape as the original kernels for the convolutional layer being replaced, with the same number33 ofchannels. The reviewer then suggests we analyze other non-linear activation functions to compute routing weights r(x).

We thank the reviewers for their thoughtful and constructive feedback. The reviewer notes that the proposed conditional convolution method is novel and shows promising results. The reviewer suggests that results would be more convincing if tested on more network structures. The reviewer suggests we compare our approach with squeeze-and-excite (SE). A1 performance, which includes SE layers.