PixelCNN.md

Pixel Convolutional Neural Network (PixelCNN)

Parameters

Argument	Description	Default	Choices
--train	Train model	False
--sample	Sample from model	False
--outlier_detection	Outlier detection	False
--dataset	Dataset name	mnist	mnist, cifar10, fashionmnist, chestmnist, octmnist, tissuemnist, pneumoniamnist, svhn, tinyimagenet, cifar100, places365, dtd, imagenet
--no_wandb	Disable Wandb	False
--out_dataset	Outlier dataset name	fashionmnist	mnist, cifar10, fashionmnist, chestmnist, octmnist, tissuemnist, pneumoniamnist, svhn, tinyimagenet, cifar100, places365, dtd,imagenet
--batch_size	Batch size	128
--n_epochs	Number of epochs	100
--lr	Learning rate	1e-3
--gamma	Gamma for the lr scheduler	0.99
--sample_and_save_freq	Sample and save frequency	5
--hidden_channels	Number of channels for the convolutional layers	64
--checkpoint	Checkpoint path	None
--num_workers	Number of workers for Dataloader	0

Training

The PixelCNN can be trained with:

python P-CNN.py --train

Sampling

For sampling you must provide the checkpoint:

python P-CNN.py --sample --checkpoint ./../../models/PixelCNN/PixelCNN_mnist.pt

Outlier Detection

Outlier Detection is performed by using the NLL scores generated by the model:

python P-CNN.py --outlier_detection --checkpoint ./../../models/PixelCNN/PixelCNN_mnist.pt