Table 3 Ablation study on the structural encoder

From: DCADif: decoupled conditional adaptive time-dynamic fusion diffusion inpainting of traditional Chinese mural paintings

CLIP Model

PSNR ↑

SSIM ↑

LPIPS ↓

w/o Frozen CLIP

26.942

0.926

0.013

Frozen CLIP (Ours)

26.942

0.925

0.013

  1. We compare the performance with a frozen CLIP encoder versus without it (W/O frozen CLIP). The test was conducted on masks with a 20%-30% corruption ratio. *Optimal results are displayed in bold.
Back to article page