https://github.com/eddytheco/gan
https://github.com/eddytheco/gan
Last synced: 2 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/eddytheco/gan
- Owner: EddyTheCo
- Created: 2022-10-28T12:56:15.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2022-11-07T13:15:08.000Z (over 2 years ago)
- Last Synced: 2025-02-07T08:34:38.630Z (4 months ago)
- Language: C++
- Size: 14.6 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Generative Adversarial Network application based on libtorch
This repository is a "general" application for implementing Generative Adversarial Network using libtorch.
It uses [yaml](https://github.com/jbeder/yaml-cpp) [input files](INPUT) for configuration of the trainning process.
## Dependencies
* libtorch from [Pytorch](https://pytorch.org/)
* libpng from [libpng](http://www.libpng.org/pub/png/libpng.html)
## CMake variables for configuration
|name|values|description|
|----|------|-----------|
|DISCRIMINATOR|MPS,TTN,DGAN...|The torch module to use as discriminator. One can create custom modules and add it to the namespace custom_models.|
|GENERATOR|GGAN...|The torch module to use as generator. One can create custom modules and add it to the namespace custom_models.|
|DATASET|FP2,IRIS,CMNIST,MNIST|The torch dataset to use for training the generator. One can create custom datasets and add it to the namespace custom_models::datasets.|
|TRAIN|ON,OFF|Perform training on the model.|
|TEST|ON,OFF|Perform testing on the model.|
### Note on this
Custom modules and custom datasets must have public functions of the form
|Object|Public function|
|-------------|------------------------------------|
|Custom_Module|Custom_Module(YAML::Node config)|
|Custom_Module|int64_t get_numel(void)|
|Custom_Dataset|Custom_DatasetImpl(const std::string& root, Mode mode = Mode::kTrain)|
### Testing the generator
The code test the generator by producing a image from the resulting tensor. The later tensor is the result of applying the generator to random data. This only work for the case of generators that produce images with certain restrictions.
## Install and Execute
### Build and install
```
git clone [email protected]:EddyTheCo/GAN.git GAN
cd GAN
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=install -DTEST=ON -DTRAIN=ON -DDISCRIMINATOR=DGAN -DGENERATOR=GGAN -DDATASET=MNIST -DCUSTOM_MODULES="DGAN;GGAN" ../
cmake --build . --target install -- -j4
```
### Execute
Set the different application parameter on the gan_mnist.yaml.
```
./install/bin/GAN install/INPUT/gan_mnist.yaml
```