PyRetri

Benyi Hu, Ren-Jie Song, Xiu-Shen Wei*, Yazhou Yao, Xian-Sheng Hua, Yuehu Liu

Corresponding author: Xiu-Shen Wei

Introduction

PyRetri (pronounced as [ˈperɪˈtriː]) is a unified deep learning based unsupervised image retrieval toolbox built on PyTorch, which is designed for researchers and engineers.

Major Features

PyRetri is a versatile deep learning based unsupervised image retrieval toolbox designed with simplicity and flexibility in mind.

• Modular Design: We decompose the deep learning based unsupervised image retrieval into several stages and users can easily construct an image retrieval pipeline by selecting and combining different modules.
• Flexible Loading: The toolbox is able to adapt to load several types of model parameters, including parameters with the same keys and shape, parameters with different keys, and parameters with the same keys but different shapes.
• Support of Multiple Methods: The toolbox directly supports several popluar methods designed for deep learning based unsupervised image retrieval, which is also suitable for person re-identification.
• Configuration Search Tool: We provide the configuration search tool to help users to find the optimal retrieval configuration with various hyper-parameters.

Supported Methods

The toolbox supports popluar and prominent methods of unsupervised image retrieval and users can also design and add their own modules.

• Pre-processing
  • DirectResize, PadResize, ShorterResize
  • CenterCrop, TenCrop
  • TwoFlip
  • ToTensor, ToCaffeTensor
  • Normalize
• Feature Representation
  • Global Average Pooling (GAP), Global Max Pooling (GMP)
  • R-MAC, SPoC, CroW, GeM, SCDA, PWA
  • PCB
• Post-processing
  • SVD, PCA
  • DBA
  • QE, K-reciprocal

Updates

06 May 2020: Our paper is online, you are welcomed to download the paper at here.

29 April 2020: Our PyRetri is launched!

Installation

Please refer to INSTALL.md for installation and dataset preparation.

Get Started

Please see GETTING_STARTED.md for the basic usage of PyRetri.

Model Zoo

Results and models are available in MODEL_ZOO.md.

License

This project is released under the Apache 2.0 license.

Citation

If you use this toolbox in your research, please cite this paper.

@article{PyRetri,
  title = {PyRetri: A PyTorch-based Library for Unsupervised Image Retrieval by Deep Convolutional Neural Networks},
  author = {Hu, Benyi and Song, Ren-Jie and Wei, Xiu-Shen and Yao, Yazhou and Hua, Xian-Sheng and Liu, Yuehu},
  journal= {arXiv preprint arXiv:2005.02154},
  year = {2020}
}

Contacts

If you have any questions about our work, please do not hesitate to contact us by emails.

Xiu-Shen Wei: [email protected]

Benyi Hu: [email protected]

Renjie Song: [email protected]