Fork of Tensorpack to make breaking performance improvements to the Mask RCNN example. Training is approximately 2x faster than the original implementation on AWS. - View it on GitHub