MultiShadow: Shadow Synthesis for Multiple Virtual Objects
Y. Ding, Qi Wan, Ziyuan Liu, Jie Wang · 2022
Image synthesis adds synthetic objects to existing images and makes them visually hard to distinguish. In order to make the synthetic objects more realistic, shadows are also required to be generated in the synthesis process. However, most existing shadow synthetic methods work poorly when there is no fixed lighting source or there are multiple objects that already existed in the original image. Hence in this paper, we propose a novel method called MultiShadow, which generates realistic shadows for virtual objects with better recognition of illumination information based on the existing objects in the original image. The experiments on benchmark datasets show that MultiShadow outperforms the state-of-the-art.