Detect Anything in Graphic Design: Element-Level Rewards for Autoregressive Detection
EleRPO is proposed, which extends GRPO from sequence-level supervision to element-level optimization, and provides fine-grained training signals that capture how each detected element contributes to overall detection quality, and works synergistically with compositional order to improve detection performance.
Jiang-Ning Zhu, Bo-Wen Li, Shenyu Qiao et al.
· 1 citation