FactorVAE

Pytorch implementation of FactorVAE proposed in Disentangling by Factorising, Kim et al.([http://arxiv.org/abs/1802.05983])

Dependencies

python 3.6.4
pytorch 0.4.0 (or check pytorch-0.3.1 branch for pytorch 0.3.1)
visdom
tqdm

Datasets

1. 2D Shapes(dsprites) Dataset

   sh scripts/prepare_data.sh dsprites

2. 3D Chairs Dataset

   sh scripts/prepare_data.sh 3DChairs

3. CelebA Dataset([download])

   
   # first download img_align_celeba.zip and put in data directory like below
   └── data
   └── img_align_celeba.zip

then run scrip file

sh scripts/prepare_data.sh CelebA

then data directory structure will be like below<br>

. └── data └── dsprites-dataset └── dsprites_ndarray_co1sh3sc6or40x32y32_64x64.npz ├── 3DChairs └── images ├── 1_xxx.png ├── 2_xxx.png ├── ... ├── CelebA └── img_align_celeba ├── 000001.jpg ├── 000002.jpg ├── ... └── 202599.jpg └── ...

NOTE: I recommend to preprocess image files(e.g. resizing) BEFORE training and avoid preprocessing on-the-fly.
<br>

### Usage
initialize visdom

python -m visdom.server

you can reproduce results below as follows

e.g. sh scripts/run_celeba.sh $RUN_NAME sh scripts/run_dsprites_gamma6p4.sh $RUN_NAME sh scripts/run_dsprites_gamma10.sh $RUN_NAME sh scripts/run_3dchairs.sh $RUN_NAME

or you can run your own experiments by setting parameters manually

e.g. python main.py --name run_celeba --dataset celeba --gamma 6.4 --lr_VAE 1e-4 --lr_D 5e-5 --z_dim 10 ...

check training process on the visdom server

localhost:8097

<br>

### Results - 2D Shapes(dsprites) Dataset
#### Reconstruction($\gamma$=6.4)
<p align="center">
<img src=misc/2DShapes_reconstruction_gamma6p4_700000.jpg>
</p>

#### Latent Space Traverse($\gamma$=6.4)
<p align="center">
<img src=misc/2DShapes_fixed_ellipse_gamma6p4_700000.gif>
<img src=misc/2DShapes_fixed_square_gamma6p4_700000.gif>
<img src=misc/2DShapes_fixed_heart_gamma6p4_700000.gif>
<img src=misc/2DShapes_random_img_gamma6p4_700000.gif>
</p>
<br>

#### Latent Space Traverse($\gamma$=10)
<p align="center">
<img src=misc/2DShapes_fixed_ellipse_gamma10_1000000.gif>
<img src=misc/2DShapes_fixed_square_gamma10_1000000.gif>
<img src=misc/2DShapes_fixed_heart_gamma10_1000000.gif>
<img src=misc/2DShapes_random_img_gamma10_1000000.gif>
</p>

### Results - CelebA Dataset
#### Reconstruction
<p align="center">
<img src=misc/CelebA_reconstruction_850000.jpg>
</p>

#### Latent Space Traverse
<p align="center">
<img src=misc/CelebA_traverse_850000.png>
<img src=misc/CelebA_fixed_1_850000.gif>
<img src=misc/CelebA_fixed_2_850000.gif>
<img src=misc/CelebA_fixed_3_850000.gif>
<img src=misc/CelebA_fixed_4_850000.gif>
</p>
<br>

### Results - 3D Chairs Dataset
#### Reconstruction
<p align="center">
<img src=misc/3DChairs_reconstruction_1000000.jpg>
</p>

#### Latent Space Traverse
<p align="center">
<img src=misc/3DChairs_traverse_1000000.png>
</p>
<br>

### Reference
1. Disentangling by Factorising, Kim et al.([http://arxiv.org/abs/1802.05983])

[http://arxiv.org/abs/1802.05983]: http://arxiv.org/abs/1802.05983
[download]: http://mmlab.ie.cuhk.edu.hk/projects/CelebA.html