Implements various 2d masked convolutions. Parameters ---------- mask_type : string Defines the type of mask to use. One of

Masked convolution with RGB channel splitting. Parameters ---------- mask_type : string One of 'A', 'B', 'V' or 'H'. in

Class used to generate masks. Can be used to create masks during training or to build various masks for generation. Parameters ----------

CelebA dataset. Parameters ---------- path_to_data : string Path to CelebA images. subsample : int Only load every |

Gated PixelCNN model for single channel images. Parameters ---------- img_size : tuple of ints Shape of input image. E.g. (1, 32,

Gated PixelCNN model for RGB images. Parameters ---------- img_size : tuple of ints Shape of input image. E.g. (3, 32, 32) n

Pixel Constrained CNN model. Parameters ---------- prior_net : pixconcnn.models.gated_pixelcnn.GatedPixelCNN(RGB) instance Model

Class used to train Pixel Constrained CNN models. Parameters ---------- model : pixconcnn.models.pixel_constrained.PixelConstrained insta

ResNet (with regular unmasked convolutions) mapping an conditional pixel inputs to logits. Parameters ---------- img_size : tuple of

Residual block (note that number of in_channels and out_channels must be the same). Parameters ---------- in_channels : int

Class used to train PixelCNN models without conditioning. Parameters ---------- model : pixconcnn.models.gated_pixelcnn.GatedPixelCNN(RGB

Abstract class defining PixelCNN sampling which is the same for both single channel and RGB models.