training on my own data.

[robbiewongBD]

Dear KupynOrest, As I train DeblurGAN on my own data with textures and fine details degraded by linear motion kernel, the results show that the textures and fine details cannot be well restored, which might be caused by grid artifacts introduced by the method or my poor configuration of training. I'll appreciate it if you could give more details and strategies on your training and implemention, for I simply cannot reproduce the results on your paper.

[McDo]

maybe you could replace transposed convolutions with nearest-neighbor upsampling to address checkerboard artifacts.