resample: None, 'down', or 'up'

resample: None, 'down', or 'up'

This is messy, but basically it chooses between batchnorm, layernorm, their conditional variants, or nothing, depending on the value of `name` an

load variable from npy to build the VGG :param rgb: rgb image [batch, height, width, 3] values scaled [0, 1]

conditional batchnorm (dumoulin et al 2016) for BCHW conv filtermaps

inputs: tensor of shape (batch size, num channels, width) mask_type: one of None, 'a', 'b' returns: tensor of shape (batch size, num cha

inputs: tensor of shape (batch size, num channels, height, width) mask_type: one of None, 'a', 'b' returns: tensor of shape (batch size,

inputs: tensor of shape (batch size, height, width, input_dim) returns: tensor of shape (batch size, 2*height, 2*width, output_dim)

Choose which generator and discriminator architecture to use by uncommenting one of these lines.

initialization: None, `lecun`, 'glorot', `he`, 'glorot_he', `orthogonal`, `("uniform", range)`

resample: None, 'down', or 'up'

Internal function forces updates moving_vars if is_training.

load variable from npy to build the VGG :param rgb: rgb image [batch, height, width, 3] values scaled [0, 1]

load variable from npy to build the VGG :param rgb: rgb image [batch, height, width, 3] values scaled [0, 1]