Awesome

On Calibrating Diffusion Probabilistic Models

The official code for the paper On Calibrating Diffusion Probabilistic Models.

We propose a straightforward method for calibrating diffusion probabilistic models that reduces the values of SM objectives and increases model likelihood lower bounds.

Acknowledgement

The codes are modifed based on the DPM-solver and EDM.

Reproducing CIFAR-10 results on image generation and FID

The command for computing the FID of baseline methods (without calibration):

python main.py --config cifar10.yml \
    --exp=experiments/cifar10 \
    --sample --fid \
    --timesteps=20 \
    --eta 0 --ni \
    --skip_type=logSNR \
    --sample_type=dpm_solver \
    --start_time=1e-4 \
    --dpm_solver_fast -i baseline

The command for computing the FID of our methods (with calibration):

python main.py --config cifar10.yml \
    --exp=experiments/cifar10 \
    --sample --fid \
    --timesteps=20 \
    --eta 0 --ni \
    --skip_type=logSNR \
    --sample_type=dpm_solver \
    --start_time=1e-4 \
    --dpm_solver_fast -i our --score_mean 

Reproducing CelebA results on image generation and FID

The command for computing the FID of baseline methods (without calibration):

python main.py --config celeba.yml \
    --exp=experiments/celeba \
    --sample --fid \
    --timesteps=50 \
    --eta 0 --ni \
    --skip_type=logSNR \
    --sample_type=dpm_solver \
    --start_time=1e-4 \
    --dpm_solver_fast -i baseline

The command for computing the FID of our methods (with calibration):

python main.py --config celeba.yml \
    --exp=experiments/celeba \
    --sample --fid \
    --timesteps=50 \
    --eta 0 --ni \
    --skip_type=logSNR \
    --sample_type=dpm_solver \
    --start_time=1e-4 \
    --dpm_solver_fast -i our --score_mean 

Estimating SDE likelihood

The command for running on CIFAR-10:

python main.py --config cifar10.yml \
    --exp=experiments/cifar10 \
    --sample --eta 0 \
    --ni --start_time=1e-4 \
    -i temp --likelihood sde

The command for running on CelebA:

python main.py --config celeba.yml \
    --exp=experiments/celeba \
    --sample --eta 0 \
    --ni --start_time=1e-4 \
    -i temp --likelihood sde

Estimating the average estimated score with EDM

cd edm/;

# CIFAR-10
python torch.distributed.run --master_port 12315 --nproc_per_node=1 generate.py --outdir=generations/cifar10/temp --seeds=0-49999 --subdirs --method our --network=https://nvlabs-fi-cdn.nvidia.com/edm/pretrained/edm-cifar10-32x32-uncond-vp.pkl

# ImageNet
python torch.distributed.run --master_port 12311 --nproc_per_node=1 generate.py --outdir=generations/imagenet/temp --seeds=0-49999 --subdirs --steps=256 --S_churn=40 --S_min=0.05 --S_max=50 --S_noise=1.003 --method our --network=https://nvlabs-fi-cdn.nvidia.com/edm/pretrained/edm-imagenet-64x64-cond-adm.pkl

The commands for running on FFHQ and AFHQv2 are similar.