Closed brisyramshere closed 4 years ago
Hi, for the decoder's structure we mostly follow the MUNIT paper which uses LayerNorm to decouple the dependency of batch sizes in BatchNorm layers. With BathNorm, the performance usually drops during inference when the batch size is smaller.
Hi, for the decoder's structure we mostly follow the MUNIT paper which uses LayerNorm to decouple the dependency of batch sizes in BatchNorm layers. With BathNorm, the performance usually drops during inference when the batch size is smaller.