FineGAN: Unsupervised Hierarchical Disentanglement for Fine-grained Object Generation and Discovery
Clone or download
Latest commit 0f67435 Dec 5, 2018
Type Name Latest commit message Commit time
Failed to load latest commit information.
code Updated Dec 6, 2018
data Initial commit Dec 5, 2018
files Initial commit Dec 5, 2018
models Initial commit Dec 5, 2018 Added hard negative training Dec 6, 2018


Pytorch implementation for learning to synthesize images in a hierarchical, stagewise manner by disentangling background, object shape and object appearance.

FineGAN: Unsupervised Hierarchical Disentanglement for Fine-grained Object Generation and Discovery

Krishna Kumar Singh*, Utkarsh Ojha*, Yong Jae Lee
[project page] [arxiv] [demo video]



  • Linux
  • Python 2.7
  • Pytorch 0.4.1
  • TensorboardX 1.2

Getting started

Clone the repository

git clone
cd finegan

Setting up the data

Note: You only need to download the data if you wish to train your own model.

Download the formatted CUB data from this link and extract it inside the data directory

cd data
cd ..

Downloading pretrained models

Pretrained generator models for CUB, Stanford Dogs are available at this link. Download and extract them in the models directory.

cd models
cd ../code/

Evaluating the model

In cfg/eval.yml:

  • Specify the model path in TRAIN.NET_G.
  • Specify the output directory to save the generated images in SAVE_DIR.
  • Specify the number of super and fine-grained categories in SUPER_CATEGORIES and FINE_GRAINED_CATEGORIES according to our paper.
  • Specify the option for using 'tied' latent codes in TIED_CODES:
    • if True, specify the child code in TEST_CHILD_CLASS. The background and parent codes are derived through the child code in this case.
    • if False, i.e. no relationship between parent, child or background code, specify each of them in TEST_PARENT_CLASS, TEST_CHILD_CLASS and TEST_BACKGROUND_CLASS respectively.
  • Run python --cfg cfg/eval.yml --gpu 0

Training your own model

In cfg/train.yml:

  • Specify the dataset location in DATA_DIR.
    • NOTE: If you wish to train this on your own (different) dataset, please make sure it is formatted in a way similar to the CUB dataset that we've provided.
  • Specify the number of super and fine-grained categories that you wish for FineGAN to discover, in SUPER_CATEGORIES and FINE_GRAINED_CATEGORIES.
  • Specify the training hyperparameters in TRAIN.
  • Run python --cfg cfg/train.yml --gpu 0

Sample generation results of FineGAN

1. Stage wise image generation results

2. Grouping among the generated images (child).


If you find this code useful in your research, consider citing our work:

  title = {FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery},
  author = {Krishna Kumar Singh, Utkarsh Ojha, and Yong Jae Lee},
  booktitle = {Arxiv},
  year = {2018}


We thank the authors of StackGAN++: Realistic Image Synthesis with Stacked Generative Adversarial Networks for releasing their source code.


For any questions regarding our paper or code, contact Krishna Kumar Singh and Utkarsh Ojha.