A PyTorch Dual Path Network implementation with pretrained MXNet weight conversion

I’ve posted a PyTorch implementation of Dual Path Networks (DPN) that works with the pretrained MXNet weights of the original implementation by cypw and achieves comparable validation results. A summary of results is included in the README with Torchvision models and two of Cadene’s Inception ports included as reference.

I’ve also included a really simple inference example script as I’ve seen others asking…