Training one-step generative models by estimating the score of mixtures of real and fake samples across noise levels yields stable training and competitive FID on CIFAR-10 and ImageNet 64x64.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Score-of-Mixture Training: Training One-Step Generative Models Made Simple via Score Estimation of Mixture Distributions
Training one-step generative models by estimating the score of mixtures of real and fake samples across noise levels yields stable training and competitive FID on CIFAR-10 and ImageNet 64x64.