BAgger: Backwards Aggregation for Mitigating Drift in Autoregressive Video Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Po, Ryan, Chan, Eric Ryan, Chen, Changan, Wetzstein, Gordon
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917144310054912
author Po, Ryan
Chan, Eric Ryan
Chen, Changan
Wetzstein, Gordon
author_facet Po, Ryan
Chan, Eric Ryan
Chen, Changan
Wetzstein, Gordon
contents Autoregressive video models are promising for world modeling via next-frame prediction, but they suffer from exposure bias: a mismatch between training on clean contexts and inference on self-generated frames, causing errors to compound and quality to drift over time. We introduce Backwards Aggregation (BAgger), a self-supervised scheme that constructs corrective trajectories from the model's own rollouts, teaching it to recover from its mistakes. Unlike prior approaches that rely on few-step distillation and distribution-matching losses, which can hurt quality and diversity, BAgger trains with standard score or flow matching objectives, avoiding large teachers and long-chain backpropagation through time. We instantiate BAgger on causal diffusion transformers and evaluate on text-to-video, video extension, and multi-prompt generation, observing more stable long-horizon motion and better visual consistency with reduced drift.
format Preprint
id arxiv_https___arxiv_org_abs_2512_12080
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle BAgger: Backwards Aggregation for Mitigating Drift in Autoregressive Video Diffusion Models
Po, Ryan
Chan, Eric Ryan
Chen, Changan
Wetzstein, Gordon
Computer Vision and Pattern Recognition
Machine Learning
Autoregressive video models are promising for world modeling via next-frame prediction, but they suffer from exposure bias: a mismatch between training on clean contexts and inference on self-generated frames, causing errors to compound and quality to drift over time. We introduce Backwards Aggregation (BAgger), a self-supervised scheme that constructs corrective trajectories from the model's own rollouts, teaching it to recover from its mistakes. Unlike prior approaches that rely on few-step distillation and distribution-matching losses, which can hurt quality and diversity, BAgger trains with standard score or flow matching objectives, avoiding large teachers and long-chain backpropagation through time. We instantiate BAgger on causal diffusion transformers and evaluate on text-to-video, video extension, and multi-prompt generation, observing more stable long-horizon motion and better visual consistency with reduced drift.
title BAgger: Backwards Aggregation for Mitigating Drift in Autoregressive Video Diffusion Models
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2512.12080