https://github.com/victordibia/tpudcgan

Train DCGAN with TPUs on Google Cloud
https://github.com/victordibia/tpudcgan

dcgan dcgan-tensorflow deep-learning gan machine-learning tpu

Last synced: about 1 month ago
JSON representation

Train DCGAN with TPUs on Google Cloud

• Host: GitHub
• URL: https://github.com/victordibia/tpudcgan
• Owner: victordibia
• Created: 2018-12-21T04:16:22.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2018-12-30T06:08:55.000Z (about 6 years ago)
• Last Synced: 2024-12-17T18:50:34.028Z (about 2 months ago)
• Topics: dcgan, dcgan-tensorflow, deep-learning, gan, machine-learning, tpu
• Language: Python
• Homepage:
• Size: 6.51 MB
• Stars: 39
• Watchers: 8
• Forks: 6
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
## Train a GAN using TPUs and Tensorflow on Google Cloud

> Screenshot below shows 64px images generated using the code provided.



This repo contains code to train an unconditional DCGAN ([Radford et al 2017](http://www.arxiv.org/pdf/1511.06434.pdf)) using TPUs on Google Cloud. It is based on the [DCGAN TPU example](https://github.com/tensorflow/tpu/tree/master/models/experimental/dcgan) by the Google Tensorflow team with the following modifications  

- Support for `64*64` and `128*128` generation: Provide two model architectures (mainly additional layers) that support generating higher resolution images (64, 128).

- Images to TFRecords: A [script](utils/convert_to_tfrecords.py) is available to convert images in a folder to TFRecords required to train the DCGAN.

- Trained models: Trained [models](models) for generating masks are available in the models folder and a [script](generate_from_model.py) for generating images is included.

## Convert Images

The `convert_to_tfrecords` script accepts arguments for data directory (`data_dir`) and output file (`output_file`). Data directory is expected to have folders which contain images directly.

```shell

python convert_to_tfrecords --data_dir=images/cifar --output_file=images/cifar/train.tfrecords --image_size=128

```

Expected

```

images

├── cifar

    ├── train

        └── train_image1.jpg

        └── train_image2.jpg

    └── test

        └── test_image1.jpg

        └── test_image2.jpg

```

## Training

- Please follow the official tensorflow tutorial on [setting up a TPU instance](https://cloud.google.com/tpu/docs/quickstart). Also see tutorial on running sample [MNIST model](https://cloud.google.com/tpu/docs/tutorials/mnist) on TPUs. 

- Clone this repo

```shell

git clone https://github.com/victordibia/tpuDCGAN

```

- Start Training

```shell

export GCS_BUCKET_NAME=  

python dcgan_main.py --tpu=$TPU_NAME --train_data_file=gs://$GCS_BUCKET_NAME/data/masks/train_masks.tfrecords   --dataset=dcgan64 --train_steps=10000 --train_steps_per_eval=500 --model_dir=gs://$GCS_BUCKET_NAME/dcgan/masks/model --test_data_file=gs://$GCS_BUCKET_NAME/data/rand/test.tfrecords

```

## Trained Models

Interested in generating masks? This repo contains two trained [models](models) (64px and 128px). You can use the generate script to generate images using any of the models. If you have your own trained DCGAN models (ckpt files) you can point the script to the model directory.

```

python generate_from_model.py --model_dir=models/masks/128/model.ckpt-15000 --image_size=128 --output_dir=models/masks/128 --random_seed=2

```