AI Paper Atlas

Research map / Video / Video generation

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

Ryan Po, Eric Ryan Chan, Changan Chen, Gordon Wetzstein

CVPR 2026 · Video › Video generation · approach: Efficient video diffusion models · 18 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add a second BAgger round (ablation)BAgger · VBench 50sImaging quality: 53.67 -> 59.98Improved
add a third BAgger round (ablation)BAgger · VBench 50sImaging quality: 59.98 -> 63.41Improved
add a second BAgger round (ablation)BAgger · VBench 50sImaging drift (lower is better): 11.83 -> 5.92Improved
add a third BAgger round (ablation)BAgger · VBench 50sImaging drift (lower is better): 5.92 -> 3.57Improved
add a third BAgger round (ablation)BAgger · VBench 50sAesthetic drift (lower is better): 3.79 -> 3.29Improved
use a single BAgger round instead of seed-only training (ablation)BAgger Round 1 · VBench 50sImaging quality: 59.97 -> 53.67Hurt
use a single BAgger round instead of seed-only training (ablation)BAgger Round 1 · VBench 50sImaging drift (lower is better): 7.34 -> 11.83Hurt
add three BAgger rounds on top of Diffusion Forcing seed training (ablation)Diffusion Forcing · VBench 50sAesthetic drift (lower is better): 5.12 -> 3.29Improved
add three BAgger rounds on top of Diffusion Forcing seed training (ablation)Diffusion Forcing · VBench 50sDynamic degree: 81.25 -> 76.56Hurt
add three BAgger rounds on top of Diffusion Forcing seed training (ablation)Diffusion Forcing · VBench 50sSubject consistency: 80.70 -> 84.05Improved
add noise to context frames at test time (sigma_test 0.2) (ablation)Diffusion Forcing · VBench 50sImaging drift (lower is better): 7.34 -> 12.27Hurt
use History Guidance instead of plain Diffusion Forcing (ablation)Diffusion Forcing · VBench 50sDynamic degree: 81.25 -> 59.38Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy