AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Group Critical-token Policy Optimization for Autoregressive Image Generation

Guohui Zhang, Hu Yu, Xiaoxiao Ma, JingHao Zhang, Yaning Pan, Mingde Yao, Jie Xiao, Linjiang Huang, Jie Huang, Feng Zhao

ICLR 2026 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images

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
optimize only 30% critical tokens with GCPO instead of all tokens with GRPOJanus-Pro-7B · GenEvaloverall: 0.87 -> 0.90Improved
GCPO instead of GRPOJanus-Pro-1B · GenEvaloverall: 0.84 -> 0.85Improved
GCPO instead of GRPOLlamaGen (HPS reward) · GenEvaloverall: 0.39 -> 0.42Improved
GCPO instead of GRPO on out-of-distribution T2I-CompBenchJanus-Pro-7B · T2I-CompBench Shapescore: 0.3999 -> 0.5173Improved
GCPO instead of GRPO on Janus-Pro-1B, T2I-CompBench ComplexJanus-Pro-1B · T2I-CompBench Complexscore: 0.3159 -> 0.3059Hurt
GCPO instead of GRPO with HPS rewardJanus-Pro-7B · DrawBench PickScorePickScore: 22.12 -> 22.10No clear effect
add dynamic advantage weight on top of three critical-token types (ablation)GCPO (Janus-Pro-1B) · GenEvaloverall: 0.83 -> 0.85Improved
combine initial, high-entropy-gradient and low-similarity tokens instead of a single type (ablation)GCPO (Janus-Pro-1B) · GenEvaloverall: 0.82 -> 0.83 (best single type vs all three)Improved
select critical tokens instead of random tokens at the same budget (ablation)GRPO · GenEvalscoreImproved
train GRPO on the remaining 70% of tokens rather than critical 30% (ablation)GRPO · GenEvalscoreHurt
raise critical token selection above 10% of tokens per type (ablation)GRPO · GenEvalscore: +2.81 within 10%, +0.95 beyondNo clear effect

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