AI Paper Atlas

Research map / Video / Video generation

VGGRPO: Towards World-Consistent Video Generation with 4D Latent Reward

Zhaochong An, Orest Kupyn, Théo Uscidda, Andrea Colaco, Karan Ahuja, Serge Belongie, Mar Gonzalez Franco, Marta Gazulla

ECCV 2026 · Video › Video generation · approach: Autoregressive long video generation · 24 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 geometry reprojection reward to motion-smoothness reward (ablation)VGGRPO · Static+dynamic avg (Wan2.2-5B)VideoReward visual quality win rate: 55.60 -> 59.57Improved
add geometry reprojection reward to motion-smoothness reward (ablation)VGGRPO · Static epipolar errorSampson epipolar error: 0.104 -> 0.093Improved
add geometry reprojection reward to motion-smoothness reward (ablation)VGGRPO · Static+dynamic avg (Wan2.2-5B)VideoReward motion quality win rate: 63.40 -> 67.21Improved
use Any4D instead of VGGT as geometry foundation model (ablation)VGGRPO · Static+dynamic avg (Wan2.2-5B)VideoReward visual quality win rate: 54.96 -> 59.57Improved
use Any4D instead of VGGT as geometry foundation model (ablation)VGGRPO · Static epipolar errorSampson epipolar error: 0.090 vs 0.093No clear effect
compute rewards in latent space instead of decoded RGB (ablation)VGGRPO · Reward computation timetime (s): 54.73 -> 41.33Improved
compute rewards in latent space instead of decoded RGB (ablation)VGGRPO · Reward computation memorypeak GPU memory (GB): 76.80 -> 68.57Improved
add test-time latent reward guidance, training-freeWan2.2-5B base · Static epipolar errorSampson epipolar error: 0.142 -> 0.136Improved
add test-time latent reward guidance, training-freeWan2.2-5B base · Sampling timetime (s): 44.35 -> 62.60Hurt
apply VGGRPO post-trainingWan2.2-5B · Standard VBench captionsImaging quality: 0.6733 -> 0.6861Improved
apply VGGRPO post-trainingWan2.2-5B · Standard VBench captionsDynamic degree: 0.4237 -> 0.3962Hurt
post-train with Epipolar-DPO or VideoGPA instead of nothingWan2.2-5B · Standard VBench captionsImaging quality: 0.6733 -> 0.6353Hurt

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