RealNVP.md

RealNVP

This work implements the Real-valued Non-Volume Preserving (RealNVP) transformations, a set of powerful invertible and learnable transformations, resulting in an unsupervised learning algorithm with exact log-likelihood computation, exact sampling, exact inference of latent variables, and an interpretable latent space.

Parameters

Argument	Description	Default	Choices
--train	Train model	False
--sample	Sample 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
--weight_decay	Weight decay	1e-5
--max_grad_norm	Max grad norm	100.0
--sample_and_save_freq	Sample and save frequency	5
--num_scales	Number of scales	2
--mid_channels	Mid channels	64
--num_blocks	Number of blocks	8
--checkpoint	Checkpoint path	None
--num_workers	Number of workers for Dataloader	0

You can find out more about the parameters by checking util.py or by running the following command on the example script:

python RNVP.py --help

Training

You can train this model with the following command:

python RNVP.py --train --dataset octmnist

Sampling

To sample, please provide the checkpoint:

python RNVP.py --sample --dataset octmnist --checkpoint ./../../models/RealNVP/RealNVP_octmnist.pt

Outlier Detection

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

python RNVP.py --outlier_detection --dataset octmnist --out_dataset mnist --checkpoint ./../../models/RealNVP/RealNVP_octmnist.pt