I'm wondering whether there's a Sparse ResNet implementation, which is also referred in the paper. I've tried to implement ResNet50 in the similar way as your DenseNet example, but only achieves 90% accuracy on Cifar10. Is there an additional transition layer required between each residual block?
I'm wondering whether there's a Sparse ResNet implementation, which is also referred in the paper. I've tried to implement ResNet50 in the similar way as your DenseNet example, but only achieves 90% accuracy on Cifar10. Is there an additional transition layer required between each residual block?
Looking forward to your reply.