search

hezhangsprinter / DCPDN

Densely Connected Pyramid Dehazing Network (CVPR'2018)
401 stars 112 forks source link

inconsistent tensor sizes in"dout8_out7 = torch.cat([dout8, out7], 1)" #14

Open my1347 opened 6 years ago

my1347 commented 6 years ago

image

my1347 commented 6 years ago

Using torch 0.3.1 image

my1347 commented 6 years ago

It seems that input image size must be less than 512??

hezhangsprinter commented 6 years ago

I guess 1024 should also work. However, since the network is trained on 512X512, the receptive field will be a issue. To get good results, please crop or resize to 512X512 first.