notesum.ai

Published at December 5

VASCAR: Content-Aware Layout Generation via Visual-Aware Self-Correction

cs.CV

Released Date: December 5, 2024

Authors: Jiahao Zhang¹, Ryota Yoshihashi, Shunsuke Kitada, Atsuki Osanai, Yuta Nakashima

Aff.: ¹University

Arxiv: http://arxiv.org/pdf/2412.04237v1

Refer to caption

	PKU unannotated				CGL unannotated
Method	Occ $\downarrow$	Rea $\downarrow$	Und $\uparrow$	Ove $\downarrow$	Occ $\downarrow$	Rea $\downarrow$	Und $\uparrow$	Ove $\downarrow$
CGL-GAN [59]	0.191	0.0312	0.32	0.069	0.481	0.0568	0.26	0.269
DS-GAN [17]	0.180	0.0301	0.52	0.026	0.435	0.0563	0.29	0.071
ICVT [6]	0.189	0.0317	0.48	0.292	0.446	0.0425	0.67	0.301
LayoutDM [19]	0.165	0.0285	0.38	0.201	0.421	0.0506	0.49	0.069
Autoreg [16]	0.154	0.0274	0.35	0.022	0.384	0.0427	0.76	0.058
RALF [16]	\ul0.133	\ul0.0231	\ul0.87	\ul0.018	\ul0.336	\ul0.0397	\ul0.93	\ul0.027
VASCAR (Ours)	0.132	0.0151	0.98	0.0002	0.289	0.0300	0.98	0.0007