作者位次 #2/8ACL 2026Poster· Natural Language Processing
From Competition to Synergy: Unlocking Reinforcement Learning for Subject-Driven Image Generation
Ziwei Huang · Ying Shu · Fanghao · Quanyu Long · Wenya Wang · Qiushi Guo · Tiezheng Ge · Leilei Gan
摘要
Subject-driven image generation models face a fundamental trade-off between identity preservation (fidelity) and prompt adherence (editability). While online reinforcement learning (RL), specifically GPRO, offers a promising solution, we find that a naive application of GRPO leads to competitive...