CapsGAN

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

★ 57 Open GitHub ↗

Capsule GAN: Unsupervised representation learning with CapsNet based Generative Adversarial Networks

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# CapsGAN

*_Work In Progress: Some major code refactoring is underway and this branch is unstable a.t.m_ *.

Unsupervised representation learning with CapsNet based Generative Adversarial Networks

Arxiv paper link [here](https://arxiv.org/abs/1806.03968)

## Description

Coalesces Wassertein Distance algorithm based generative adversarial networks with deep convolutinal network based discriminator replaced by CapsNet architecture.

## Prerequisits and Requirements

* Python > 3.6
* PyTorch
* TorchVision
* TorchNet
* TQDM
* Visdom [_optional_]

** For training, an NVIDIA TITAN XP GPU was used. __Using CUDA with GPU is strongly recommended__. CPU is supported but speed of training will be very slow.

## Datasets

* MNIST
* CIFAR-10
* SmallNORB

### MNIST ###

__Running MNIST__

Pass in the gpu device number for e.g. `0`
```
$ python --dataset mnist --nc 1 --imageSize 32 --dataroot ./data --cuda {GPU_DEVICE_NUMBER}  --workers 4
```

### SmallNORB ###
__Architecture__

High level architecture:

![SmallNORB_Architecture](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/em-capsule-architecture.png)

__Running SmallNORB__

[SmallNORB](https://cs.nyu.edu/~ylclab/data/norb-v1.0-small/) is an ideal dataset for testing the efficacy of Capsule networks over traditional CNN based GANs.

![SmallNORB_Animals](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/real_samples_animals.png)
![SmallNORB_Humans](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/real_samples_humans.png)
![SmallNORB_Planes](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/real_samples_planes.png)
![SmallNORB_Trucks](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/real_samples_trucks.png)
![SmallNORB_Cars](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/real_samples_cars.png)

Pass in the gpu device number for e.g. `0`
```
$ python --dataset smallnorb --nc 1 --imageSize 64 --dataroot ./data/smallnorb/ --cuda {GPU_DEVICE_NUMBER}  --workers 4
```

__SmallNORB Results__

Reconstructed smallNORB cars and planes:

![Reconstructed SmallNORB Cars](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/caps_gen_car.png)
![Reconstructed SmallNORB Planes](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/smallnorb/caps_gen_plane.png)

### CIFAR10 ###
__Running CIFAR10__

Pass in the gpu device number for e.g. `0`
```bash
$ python3 main.py --nc 3 --dataset cifar10 --dataroot ./data --cuda {GPU_DEVICE_NUMBER} --workers 4 --niter [NUM_EPOCHS] 
```

## Usage

Utilizes FAIR's visdom as visulization tool. If you'd like to visualize the test and train results, run with `visualize` args.

```bash
$ sudo python3 -m visdom.server &
$ python3 main.py --visualize --cuda
```

To run with MLP as G or D, run:
```bash
$ python3 main.py --dataset cifar10 --dataroot ./data --cuda {device_num} --experiment {Name} --mlp_G --ngf 512
```

__Using CUDA__

Pass in the gpu device number for e.g. `0`
```
$ python main.py --cuda {GPU_DEVICE_NUMBER}
```

__Enable Visualization__

Start the server (probably in a screen or tmux):
```bash
python -m visdom.server -port 8097
```

Run with `--visualize` parameter
```
$ python main.py --cude {GPU_DEVICE_NUMBER} --visualize
```

## Architecture

Using DCGAN (and variants - BN, no-BN) as baseline against the CapsNet architecture.

![DCGAN](https://raw.githubusercontent.com/raeidsaqur/CapsGAN/master/img/DCGAN_architecture.png)

## Contact
Please send an email to raeidsaqur[at]cs[dot]toronto[dot]edu for questions, PRs etc.

*** Note: __Improved ReadMe is in the works!__

CapsGAN

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README