WMGM generates 128x128 images by diffusing only low-frequency wavelet coefficients and using a shared multi-scale GAN to fill in high-frequency details, improving FID and cutting sampling time and parameters versus SGM and WSGM.
Image compression via joint statistical charac- terization in the wavelet domain
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Multi-scale Generative Modeling for Fast Sampling
WMGM generates 128x128 images by diffusing only low-frequency wavelet coefficients and using a shared multi-scale GAN to fill in high-frequency details, improving FID and cutting sampling time and parameters versus SGM and WSGM.