Skip to main content

Table 3 Comparisons of methods on the FiveK test dataset

From: Stacked generative adversarial networks for image compositing

  PSNR SSIM LPIPS
Cut-and-paste 24.40db 0.9547 0.0410
MVCC [1] 27.23db 0.9487 0.0363
IM [8] 22.95db 0.8349 0.0945
DIH [3] 29.96db 0.9368 0.0401
Ours (w/o G2) 33.22db 0.9628 0.0298
Ours (w/ D1 only) 33.66db 0.9517 0.0248
Ours (w/o Shift) 32.46db 0.9504 0.0252
Ours (w/o \(\mathcal {L}_{\text {per}}\)) 33.41db 0.9512 0.0422
Ours (w/o \(\mathcal {L}_{\text {gui}}\)) 34.42db 0.9580 0.0217
Ours 34.74db 0.9692 0.0209