https://github.com/muhammedbuyukkinaci/flowergan
FlowerGAN : A DCGAN implementation in TensorFlow
https://github.com/muhammedbuyukkinaci/flowergan
dcgan dcgan-tensorflow flower flowergan gan generative-adversarial-network gpu tensorflow
Last synced: 6 months ago
JSON representation
FlowerGAN : A DCGAN implementation in TensorFlow
- Host: GitHub
- URL: https://github.com/muhammedbuyukkinaci/flowergan
- Owner: MuhammedBuyukkinaci
- Created: 2018-05-12T16:07:51.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2018-07-07T22:45:35.000Z (over 7 years ago)
- Last Synced: 2025-05-07T09:04:26.399Z (6 months ago)
- Topics: dcgan, dcgan-tensorflow, flower, flowergan, gan, generative-adversarial-network, gpu, tensorflow
- Language: Python
- Size: 89.8 MB
- Stars: 13
- Watchers: 0
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# FlowerGAN
FlowerGAN : A DCGAN implementation in TensorFlow in Python 3 on 102 Category Flower Dataset.
# Dependencies
```pip install -r requirements.txt```
or
```pip3 install -r requirements.txt```
# Training
```git clone https://github.com/MuhammedBuyukkinaci/FlowerGAN.git```
```cd ./FlowerGAN```
```python FlowerGAN.py ```
# Notebook
You can download .ipynb file from [here](https://github.com/MuhammedBuyukkinaci/My-Jupyter-Files-1/blob/master/FLOWER_dcgan.ipynb).
```jupyter lab ``` or ```jupyter notebook ```
# Data
No MNIST or CIFAR-10.
This is a repository containing datasets of 8189 flower pictures belonging to 102 different categories. We aren't interested in
categories because GAN's is an UNSUPERVISED Machine Learning model.
Download .tgz extension version from [here](
http://www.robots.ox.ac.uk/~vgg/data/flowers/102/) or .npy extension version from [here](
https://www.dropbox.com/s/gmu2cxgjktcnw40/flower_photos.npy?dl=0). It is about 95 MB.
If you downloaded the dataset, extract files from 102flowers.tgz .Then put it in FlowerGAN folder.
If you download .npy file from Dropbox, put flower_photos.npy in FlowerGAN folder.
# GPU
I trained on GTX 1050. 1 epoch lasted 6-7 minutes. I left my laptop overnight and obtained outputs in the morning.
If you don't want to wait for one month, use a GPU.
# Architecture
Images are resized to (64,64,3) . The architecture is below:
# Generated Photos
Each picture contains 16 generated photos. I trained it 100 epochs and obtained outputs 1 in 2 epochs.
