caojiezhang / LCCGAN

Code for “Adversarial Learning with Local Coordinate Coding”

#LCC-GANs (ICML2018)

Pytorch implementation for “Adversarial Learning with Local Coordinate Coding”.

Architecture of LCCGAN

• AutoEncoder (AE) learns embeddings on the latent manifold.
• Local Coordinate Coding (LCC) learns local coordinate systems.
• The LCC sampling method is conducted on the latent manifold.

Gometric Views of LCC Sampling

• With the help of LCC, we obtain local coordinate systems for sampling on the latent manifold.
• Using the local coordinate systems, LCC-GANs always sample some meaningful points to generate new images with different attributes.

Training Algorithm

Dependencies

python 2.7

Pytorch

Dataset

In our paper, to sample different images, we train our model on four datasets, respectively.

• Download MNIST dataset.

• Download Oxford-102 Flowers dataset.

• Download Large-scale CelebFaces Attributes (CelebA) dataset.

• Download Large-scale Scene Understanding (LSUN) dataset.

Training

• Train AEGAN on Oxford-102 Flowers dataset.

python train.py --dataset flowers --dataroot your_images_folder --batchSize 64 --imageSize 64 --cuda

• If you want to train the model on Large-scale CelebFaces Attributes (CelebA), Large-scale Scene Understanding (LSUN) or your own dataset. Just replace the hyperparameter like these:

python train.py --dataset name_o_dataset --dataroot path_of_dataset

Citation

@InProceedings{pmlr-v80-cao18a,
  title = 	 {Adversarial Learning with Local Coordinate Coding},
  author = 	 {Cao, Jiezhang and Guo, Yong and Wu, Qingyao and Shen, Chunhua and Huang, Junzhou and Tan, Mingkui},
  booktitle = 	 {Proceedings of the 35th International Conference on Machine Learning},
  pages = 	 {707--715},
  year = 	 {2018},
  editor = 	 {Dy, Jennifer and Krause, Andreas},
  volume = 	 {80},
  series = 	 {Proceedings of Machine Learning Research},
  address = 	 {Stockholmsmässan, Stockholm Sweden},
  month = 	 {10--15 Jul},
  publisher = 	 {PMLR}
}