SEMBG

Introduction

This repository is the Pytorch code for the paper [Low-Cost Self-Ensembles Based on Multi-Branch Transformation and Grouped Convolution].

These codes are examples for CIFAR100 with Wide_ResNet28-10.

Dependencies

• Python 3.6.13 (Anaconda)
• Pytorch 1.7.1
• CUDA 10.1

Run

run for training a single model:

python train_single.py

run for training SEMBG (N=3):

python train_SEMBG.py

Citation

@article{SEMBG,
  title={Low-Cost Self-Ensembles Based on Multi-Branch Transformation and Grouped Convolution},
  author ={H. {Lee} and J. -S. {Lee}},
  journal = {arXiv preprint arXiv:2408.02307}
  year={2024},
}