List your company
ResearchFormerly WizWand

Image Generation on CIFAR-10 32x32 (50K samples)

48NFESDE solver with adaptive step sizes
-30.04496.731,023.51,550.27May 28, 2021
Updated 8mo ago

Evaluation results

MethodLinks
SDE solver with adaptive step sizesModel Variant=VP deep, epsilon_rel=0.52021.054882.42
Euler-MaruyamaModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.054895.77
DDIMModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.054838.71
SDE solver with adaptive step sizesModel Variant=VP, epsilon_rel=0.52021.054972.29
Euler-MaruyamaModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.054992.99
DDIMModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.054937.24
SDE solver with adaptive step sizesModel Variant=VE deep, epsilon_rel=0.52021.0550307.32
Euler-MaruyamaModel Variant=VE deep, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.0550271.27
SDE solver with adaptive step sizesModel Variant=VE, epsilon_rel=0.52021.0552266.75
Euler-MaruyamaModel Variant=VE, NFE Strategy=Matched to Ours (epsilon_rel=0.50)2021.0552169.32
Probability Flow (ODE)Model Variant=VP2021.051423.11
Probability Flow (ODE)Model Variant=VP deep2021.051452.86
SDE solver with adaptive step sizesModel Variant=VP, epsilon_rel=0.12021.051472.95
Euler-MaruyamaModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.0514731.38
DDIMModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.0514711.53
SDE solver with adaptive step sizesModel Variant=VP deep, epsilon_rel=0.12021.051512.73
Euler-MaruyamaModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.0515131.93
DDIMModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.0515111.38
SDE solver with adaptive step sizesModel Variant=VE, epsilon_rel=0.12021.051708.85
SDE solver with adaptive step sizesModel Variant=VE deep, epsilon_rel=0.12021.0517010.15
Euler-MaruyamaModel Variant=VE, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.051705.12
Euler-MaruyamaModel Variant=VE deep, NFE Strategy=Matched to Ours (epsilon_rel=0.10)2021.051705.56
SDE solver with adaptive step sizesModel Variant=VP, epsilon_rel=0.052021.051792.59
Euler-MaruyamaModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.0517925.49
DDIMModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.051799.2
SDE solver with adaptive step sizesModel Variant=VP deep, epsilon_rel=0.052021.051802.44
Euler-MaruyamaModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.0518025.05
DDIMModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.051809.25
Probability Flow (ODE)Model Variant=VE deep2021.051815.53
Probability Flow (ODE)Model Variant=VE2021.051837.64
SDE solver with adaptive step sizesModel Variant=VE deep, epsilon_rel=0.052021.052703.4
Euler-MaruyamaModel Variant=VE deep, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.052703.76
SDE solver with adaptive step sizesModel Variant=VE, epsilon_rel=0.052021.052713.23
Euler-MaruyamaModel Variant=VE, NFE Strategy=Matched to Ours (epsilon_rel=0.05)2021.052713.48
SDE solver with adaptive step sizesModel Variant=VP, epsilon_rel=0.022021.052742.74
SDE solver with adaptive step sizesModel Variant=VP deep, epsilon_rel=0.022021.052742.6
Euler-MaruyamaModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.0527414.18
Euler-MaruyamaModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.0527413.67
DDIMModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.052745.75
DDIMModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.052745.74
SDE solver with adaptive step sizesModel Variant=VP, epsilon_rel=0.012021.053292.7
Euler-MaruyamaModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.0532910.28
DDIMModel Variant=VP, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.053294.81
SDE solver with adaptive step sizesModel Variant=VP deep, epsilon_rel=0.012021.053302.56
Euler-MaruyamaModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.0533010
DDIMModel Variant=VP deep, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.053304.76
SDE solver with adaptive step sizesModel Variant=VE deep, epsilon_rel=0.022021.054882.99
Euler-MaruyamaModel Variant=VE deep, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.054883.21
SDE solver with adaptive step sizesModel Variant=VE, epsilon_rel=0.022021.054902.87
Euler-MaruyamaModel Variant=VE, NFE Strategy=Matched to Ours (epsilon_rel=0.02)2021.054903.05
SDE solver with adaptive step sizesModel Variant=VE deep, epsilon_rel=0.012021.057363.06
Euler-MaruyamaModel Variant=VE deep, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.057363.17
SDE solver with adaptive step sizesModel Variant=VE, epsilon_rel=0.012021.057382.91
Euler-MaruyamaModel Variant=VE, NFE Strategy=Matched to Ours (epsilon_rel=0.01)2021.057382.99
Euler-MaruyamaModel Variant=VP2021.051,0002.55
Euler-MaruyamaModel Variant=VP deep2021.051,0002.49
Euler-MaruyamaModel Variant=VE2021.051,0002.98
Euler-MaruyamaModel Variant=VE deep2021.051,0003.14
DDIMModel Variant=VP2021.051,0002.86
DDIMModel Variant=VP deep2021.051,0002.69
Reverse-Diffusion & LangevinModel Variant=VP2021.051,9994.27
Reverse-Diffusion & LangevinModel Variant=VP deep2021.051,9994.69
Reverse-Diffusion & LangevinModel Variant=VE2021.051,9992.4
Reverse-Diffusion & LangevinModel Variant=VE deep2021.051,9992.21