py-began-master

This is a Boundary Equailibrium GAN(BEGAN) for image generation.paper

Dependences

• tensorflow
• opencv
• tqdm

Usage

We use CelebA dataset for training, at training state, you should run

python main.py --data_dir DATA_DIR

At the testing/interpolating step, there are three types for image interpolation:

• mode1: using one image, interpolate between it and its mirror image(horizontal flip image)
• mode2: using two images
• mode3: using CelebA images, no inputs For mode1:

python main.py --mode 1 --image1 IMAGE1_PTH

For mode2:

python main.py --mode 2 --image1 IMAGE1_PTH --image2 IMAGE2_PTH

For mode3:

python main.py --mode 3

demo

BEGAN samples(300k iterations, 64 image scale)

BEGAN interpolation(2000 interp iterations)