DualConv: Dual Convolutional Kernels for Lightweight Deep Neural Networks

This is the official PyTorch implementation for DualConv paper. If you find that this project helps your research, please consider citing the following paper:

@ARTICLE{9723436,
  author={Zhong, Jiachen and Chen, Junying and Mian, Ajmal},
  journal={IEEE Transactions on Neural Networks and Learning Systems}, 
  title={DualConv: Dual Convolutional Kernels for Lightweight Deep Neural Networks}, 
  year={2022},
  volume={},
  number={},
  pages={1-8},
  doi={10.1109/TNNLS.2022.3151138}}

Requirements

• Pytorch 1.1+
• Python 3.6+

Usage

The official implementation of our proposed DualConv and the reproduced HetConv and GroupConv are both in the /kernels folder. We provide example code for applying the lightweight convolutional kernel to the VGG-16 network on the CIFAR-10 dataset.

# Start training with: 
cd example_usage && python CIFAR-10.py --kernel dualconv

# You can manually resume the training with: 
python CIFAR-10.py --resume --lr=0.01 --kernel dualconv