A non-autoregressive diffusion framework that predicts any future frame tuple directly from context, with a motion-prediction module and an overlap-consistency training loss.
With the emergence of diffusion models [6], there have been efforts to improve prediction quality in video prediction using diffusion models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Autoregression-free video prediction using diffusion model for mitigating error propagation
A non-autoregressive diffusion framework that predicts any future frame tuple directly from context, with a motion-prediction module and an overlap-consistency training loss.