FlowerGAN : A DCGAN implementation in TensorFlow in Python 3 on 102 Category Flower Dataset.
pip install -r requirements.txt
or
pip3 install -r requirements.txt
git clone https://github.com/MuhammedBuyukkinaci/FlowerGAN.git
cd ./FlowerGAN
python FlowerGAN.py
You can download .ipynb file from here.
jupyter lab or jupyter notebook
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 or .npy extension version from here. 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.
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.
Images are resized to (64,64,3) . The architecture is below:
Each picture contains 16 generated photos. I trained it 100 epochs and obtained outputs 1 in 2 epochs.
