Unsupervised Image Classification PDF

Weijie Chen, Shiliang Pu, Di Xie, Shicai Yang, Yilu Guo, Luojun Lin. In ECCVW 2020.

Prerequisites

• python3.6
• pytorch1.1

UIC Pipeline

TLDR: UIC is a very simple self-supervised learning framework for joint image classification and representation learning. It utilizes the forward result at epoch t-1 as pseudo label to drive unsupervised training at epoch t.

Getting Started

Data Preparation

Link your own ImageNet dataset to ./data/ImageNet/train and ./data/ImageNet/val with the format of datasets.ImageFolder in Pytorch.

Unsupervised Training

$ sh ./main.sh

Linear Evaluation

$ sh ./eval_linear.sh

License

You can view in License.

Citation

If you find our code useful, please consider citing our paper:

@InProceedings{chen2020unsupervised,
  title={Unsupervised Image Classification for Deep Representation Learning},
  author={Chen, Weijie and Pu, Shiliang and Xie, Di and Yang, Shicai and Guo, Yilu and Lin, Luojun},
  booktitle={European Conference on Computer Vision Workshop},
  pages={430-446}
  year={2020},
  organization={Springer}
}

Acknowledgements

Our code is implemented based on DeepCluster.

Contact

If you are interested in internship, or applied researcher / developer positions in Hikvision Research Institute, please feel free to seed an email to chenweijie5 -at- hikvision.com.