ConditionalDDPM.md

Conditional Denoising Diffusion Probabilistic Model

The Conditional Denoising Diffusion Probabilistic Model (CDDPM) is akin to the standard DDPM, with the additional incorporation of class embeddings into the training and sampling process. These class embeddings provide the model with information about the specific class or category of the data being generated.

This model supports Accelerate for Multi-GPU and Mixed Precision Training.

Parameters

Parameter	Description	Default	Choices
--train	Train model	False
--sample	Sample model	False
--fid	Sample for FID	False
--batch_size	Batch size	128
--n_epochs	Number of epochs	100
--lr	Learning rate	0.001
--timesteps	Number of timesteps	500
--beta_start	Beta start	0.0001
--beta_end	Beta end	0.02
--dataset	Dataset name	mnist	mnist, cifar10, fashionmnist, chestmnist, octmnist, tissuemnist, pneumoniamnist, svhn, tinyimagenet, cifar100, places365, dtd, imagenet
--no_wandb	Disable Wandb	False
--ddpm	DDPM	1.0
--checkpoint	Checkpoint path	None
--sample_timesteps	Number of timesteps for sampling	500
--model_channels	Number of features	64
--num_res_blocks	Number of residual blocks per downsample	2
--attention_resolutions	Downsample rates for attention	[4]
--dropout	Dropout probability	0.0
--channel_mult	Channel multiplier for UNet levels	[1, 2, 2]
--conv_resample	Use learned convolutions for resampling	True
--dims	Signal dimensionality (1D, 2D, 3D)	2
--num_heads	Number of attention heads per layer	4
--num_head_channels	Fixed channel width per attention head	32
--use_scale_shift_norm	Use FiLM-like conditioning mechanism	False
--resblock_updown	Use residual blocks for up/downsampling	False
--use_new_attention_order	Use an alternative attention pattern	False
--n_classes	Number of classes	10
--sample_and_save_freq	Sample and save frequency	10
--drop_prob	Dropout probability	0.1
--cfg	Guidance weight	0.5
--num_workers	Number of workers for Dataloader	0
--warmup	Number of warmup epochs	10
--decay	Decay rate	1e-5
--latent	Use latent version	False
--ema_rate	Exponential moving average rate	0.999
--size	Size of input image	None

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

python CondDDPM.py --help

Training

The training process is similar to the one described in VanillaDDPM.md.

accelerate launch CondDDPM.py --train --dataset mnist --n_classes 10

Sampling

The sampling process is also similar to a typical DDPm, although the class embedding is also provided at each timestep. Conditional DDPMs can also be adjusted to sample in a deterministic manner:

python CondDDPM.py --sample --dataset mnist --n_classes 10 --checkpoint ./../../models/ConditionalDDPM/CondDDPM_mnist.pt

FID Sampling

To sample 50K images for FID computation:

python CondDDPM.py --fid --batch_size 200 --dataset cifar10 --checkpoint ./../../models/ConditionalDDPM/CondDDPM_cifar10.pt