notesum.ai

Published at November 26

Reward Incremental Learning in Text-to-Image Generation

cs.CV

cs.LG

Released Date: November 26, 2024

Authors: Maorong Wang¹, Jiafeng Mao¹, Xueting Wang¹, Toshihiko Yamasaki²

Aff.: ¹CyberAgent; ²The University of Tokyo

Arxiv: http://arxiv.org/abs/2411.17310v1

Refer to caption

Method	Target Metric	General Metrics
Method	Aesthetic score $\uparrow$	CLIP score $\uparrow$	Zero-shot FID $\downarrow$
SD V1.5	5.23	25.77	71.54
Baseline	6.07(+0.84)	22.74(-3.03)	85.53(+13.99)
RID	6.03(+0.80)	25.76(-0.01)	79.40(+7.86)