AI Paper Atlas

Research map / Video / Video generation

Rethinking Reward Signals in Video GRPO: When Scores Become Targets

Rui Li, Yuanzhi Liang, Ziqi Ni, Haibin Huang, Chi Zhang, Xuelong Li

ECCV 2026 · Video › Video generation · approach: Autoregressive long video generation · 11 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
replace DanceGRPO summed reward with TaRoS reward signaling (Qwen2.5VL-72B)Wan2.1-T2V-1.3B · VBenchTotal score: 79.72 -> 80.56Improved
replace DanceGRPO with TaRoS (VideoAlign reward)Wan2.1-T2V-1.3B · VBenchTotal score: 79.46 -> 79.84Improved
scale TaRoS reward model from Qwen2.5VL-7B to 72BTaRoS on Wan2.1-T2V-1.3B · VBenchTotal score: 79.90 -> 80.56Improved
swap DanceGRPO reward model from VideoAlign to Qwen2.5VL-72BDanceGRPO on Wan2.1-T2V-1.3B · VBenchTotal score: 79.46 -> 79.72No clear effect
apply TaRoS with Qwen2.5VL-72B to Wan2.1-14BWan2.1-T2V-14B · VBenchTotal score: 81.46 -> 82.09Improved
apply TaRoS with VideoAlign to Wan2.1-14BWan2.1-T2V-14B · VBenchTotal score: 81.46 -> 81.58No clear effect
apply TaRoS with VideoAlign to Wan2.1-14BWan2.1-T2V-14B · VBenchQuality score: 84.03 -> 83.61Hurt
apply TaRoS (Qwen2.5VL-72B) to HunyuanVideoHunyuanVideo · VBenchImage quality: 64.61 -> 68.83Improved
apply TaRoS (Qwen2.5VL-72B) to HunyuanVideoHunyuanVideo · VBenchSpatial relationship: 67.55 -> 69.84Improved
add third reward dimension to DanceGRPO (ablation)DanceGRPO · Reward ablation, iStock 1.4k promptsVideoAlign VQ: 3.493 -> 3.412Hurt
add third reward dimension to TaRoS (ablation)TaRoS · Reward ablation, iStock 1.4k promptsVideoAlign TA: -1.312 -> -0.7114Improved
use TaRoS instead of DanceGRPO with three reward dimensions (ablation)Wan2.1-T2V-1.3B · Reward ablation, iStock 1.4k promptsVideoAlign VQ: 3.412 -> 3.501Improved

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