Fast Adaptive Meta-Learning for Few-Shot Image Generation (FAML)

The official Pytorch implementation of IEEE Transactions on Multimedia 2021 Fast Adaptive Meta-Learning for Few-Shot Image Generation (FAML).

If this code is useful for you, please consider citing the paper as following detail:

@article{phaphuangwittayakul2021fast,
  title={Fast Adaptive Meta-Learning for Few-shot Image Generation},
  author={Phaphuangwittayakul, Aniwat and Guo, Yi and Ying, Fangli},
  journal={IEEE Transactions on Multimedia},
  year={2021},
  publisher={IEEE}
}

Fast Adaptive Meta-Learning (FAML) based on GAN and the encoder network is proposed in this study for few-shot image generation. The model can generate new realistic images from unseen target classes with few samples of data.

Installation

$ git clone https://github.com/phaphuang/FAML.git
$ cd FAML
$ pip install -r requirements.txt

Datasets

• MNIST: 10 balanced classes link
• Omniglot: 1623 handwritten characters from 50 different alphabets. Each of the 1623 characters was drawn online via Amazon's Mechanical Turk by 20 different people link
• VGG-Faces: 2395 categories link
• miniImageNet: 100 classes with 600 samples of 84×84 color images per class. link

Preparing datasets

The default folder of datasets are placed under data.

MNIST and Omniglot can be automatically downloaded by running the training scripts.

VggFace and miniImageNet datasets

VggFace

1. Downloaded the modified 32x32 VggFace and 64x64 VggFace datasets and their training and validation tasks.
2. Place all downloaded files in data/vgg/

miniImageNet

1. Downloaded the modified 32x32 miniImageNet and 64x64 miniImageNet datasets and their training and validation tasks.

2. Place all downloaded files in data/miniimagenet/

Training the model

$ python train.py --dataset Mnist --height 32 --length 32

View all options for different training parameters:

$ python train.py --help

Generated Outputs

MNIST Omniglot VggFace miniImageNet

Acknowledgement

Some of the codes are built upon FIGR and RaGAN. Thanks them for their great work! FAML is freely available for non-commercial use. Don't hesitate to drop e-mail if you have any problem.