The FOS score is computed by these two methods:
FOS-E:
Foreground Object Search by Distilling Composite Image Feature [arXiv][code]
Bo Zhang, Jiacheng Sui, Li Niu
Accepted by ICCV 2023.
FOS-D:
Learning a Discriminative Model for the Perception of Realism in Composite Images [arXiv][code]
Jun-Yan Zhu, Philipp Krähenbühl, Eli Shechtman and Alexei A. Efros
Accepted by ICCV 2015.
A novel FOS method via distilling composite feature (DiscoFOS). Specifically, a pretrained discriminator serves as teacher network. The student network employs two encoders to extract foreground feature and background feature. Their interaction output is enforced to match the composite image feature from the teacher network.
Training a discriminator to estimate the compatibility of composite images.