Fig. 2
From: Real-world defocus deblurring via score-based diffusion models
Defocus deblurring results of self-captured images under different models. (a)-(f) are the reconstruction results using the proposed method, the FFDNet method, the DnCNN method, the CycleGAN method, the defocus image and GT, respectively. The same rule applies to (g)-(l) and (m)-(r). (s)-(u) are enlarged images of parts (f), (l), and (r), respectively.Ours, refers to an defocus deblurring method based on a fraction-based diffusion model; FFDNet, refers to a fast and flexible denoising method based on convolutional neural networks; DnCNN, refers to a deep convolutional neural network method for image denoising; CycleGAN, refers to an image reconstruction unsupervised generative model; Input, defocus image; GT, real situation.
