This repository implements Pixel CNN in TensorFlow 2. Have Fun! -Brandon
Install this package using pip.
pip install git+git://github.com/brandontrabucco/pixelcnn.git
Create a Pixel CNN Keras Model.
model = pixelcnn.ConditionalPixelCNNPlusPlus(
256,
32,
image_height=32,
image_width=32,
conditional_height=1,
conditional_width=1,
num_preprocess_layers=5,
num_modules=3,
num_layers_per_module=6,
filters=64,
dropout_rate=0.1,
class_conditional=True,
num_classes=10)
Fetch the next batch of conditional vectors to seed the image generation process.
images = tf.zeros([12, 32, 32], dtype=tf.int32)
inputs = tf.random.uniform([12, 1, 1], maxval=10, dtype=tf.int32)
Run the model to predict image logits and use with your favorite loss function.
logits = model([images, inputs])