Pytorch-GANs

Implementation of DCGAN on CIFAR-10 dataset

• DCGAN architecture on PyTorch
• 32 x 32 x 3

Setting up the environment

• Clone the repo
• Create a virtualenvironment with python3
• Activate the environment
• Install requirements from req.txt

https://github.com/theidentity/Pytorch-GANs.git
python3 -m venv my_venv
pip3 install -r req.txt
source activate my_venv/bin/activate

Running the program

python dcgan.py --epochs=10 --gpu=0 --seed=42

Play around with the model

• Disciminator and Generator are in networks.py
• Data input and output are in data_io.py
• Custom layers are in pt_layers.py

Results

• The visualizations are in gen_imgs/dcgan/
• Current model run for 10 epochs. Train longer for more complex images
• Improvement of epochs 0 to 5 to 10 :