README.md

Content-Aware Convolution for Efficient Deep Neural Networks

Pytorch implementation for "Content-Aware Convolution for Efficient Deep Neural Networks".

Demonstration of CAC

Requirements

Python>=3.6, PyTorch==1.2.0, torchvision==0.4.0 pyhocon flame

Please follow the guide to install flame.

Datasets

We consider two benchmark classification datsets, including CIFAR-10 and ImageNet.

CIFAR-10 can be automatically downloaded by torchvision.

ImageNet needs to be manually downloaded (preferably to a SSD) following the instructions here.

Training Algorithm

Training Method

Please run the following command in cac directory:

python train.py -c config/resnet20.hocon -o results/resnet20