Autoguidance consistently improves sample quality and diversity in diffusion models, while early AJEST matches or slightly exceeds it in data efficiency but adds overhead that makes autoguidance or random selection preferable in most cases.
Data pruning in generative diffusion models, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Autoguided Online Data Curation for Diffusion Model Training
Autoguidance consistently improves sample quality and diversity in diffusion models, while early AJEST matches or slightly exceeds it in data efficiency but adds overhead that makes autoguidance or random selection preferable in most cases.