CoRe-GRPO: Consensus-driven and Region-focused RL for Human-Centric Generation in Lightweight Diffusion Models

Jun 1, 2025·
Wenjie Shu
Wenjie Shu
· 1 min read
Abstract
We propose a consensus-driven, region-focused RL method to improve human-centric generation quality in lightweight diffusion models.
Type
Publication
Under review

Under review.

Can't display the PDF inline. Click to view/download: figure2.pdf

Figure: CoRe-GRPO overview.

Diffusion Reinforcement Learning Human-Centric Generation
Wenjie Shu
Authors
Wenjie Shu (he/him)
Research Asistant