Skip to content

mibastro/VPE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Variational Prototyping-Encoder: One-Shot Learning with Prototypical Images

IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2019

We tackle an open-set graphic symbol recognition problem by one-shot classification with prototypical images as a single training example for each novel class. We take an approach to learn a generalizable embedding space for novel tasks. We propose a new approach called variational prototyping-encoder (VPE) that learns the image translation task from real-world input images to their corresponding prototypical images as a meta-task. As a result, VPE learns image similarity as well as prototypical concepts which differs from widely used metric learning based approaches. Our experiments with diverse datasets demonstrate that the proposed VPE performs favorably against competing metric learning based one-shot methods. Also, our qualitative analyses show that our meta-task induces an effective embedding space suitable for unseen data representation.

Citation

Please cite our paper in your publications if it helps your research:

@inproceedings{kim2019variational,
    title={Variational Prototyping-Encoder: One-Shot Learning with Prototypical Images},
    author={Kim, Junsik and Oh, Tae-Hyun and Lee, Seokju and Pan, Fei and Kweon, In So},
    booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
    pages={9462--9470},
    year={2019}
}

Usage

This is a PyTorch implementation of "Variational Prototyping-Encoder: One-Shot Learning with Prototypical Images" CVPR 2019. The code is based on PyTorch 1.0 and python 3.6.

  1. Clone the repository. The default folder name is 'VPE'

    git clone https://github.com/mibastro/VPE.git
  1. Download the datasets used in our paper from here. The datasets used in our paper are modified from the existing datasets. Please cite the dataset papers if you use it for your research. (Belgalogos, FlickrLogos-32, GSTRB, TT100K )

    • Organize the file structure as below.
    |__ VPE
        |__ code
        |__ db
            |__ belga
            |__ flickr32
            |__ toplogo10
            |__ GTSRB
            |__ TT100K
            |__ exp_list
    • Training and test splits are defined as text files in 'VPE/db/exp_list' folder.
  2. Set the global repository path in 'VPE/code/config.json'.

  3. Run main*.py to train and test the code.

Contact

[email protected]

About

VPE logos

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages