proteus1991/GridDehazeNet

the effect is very different

Closed this issue · 8 comments

shenhongqian-123456 commented

Hello, I train the network on my server according to your training method, the effect is very different, PSNR = 17.74, SSIM = 0.7954. What is the reason for this

proteus1991 commented

Hi,

I need more information to help you. For example, which training data are you using? How many epoches have you trained?

Thanks,

Xiaohong

shenhongqian-123456 commented

Hello, Thank you for taking the time to answer my questions
according to your paper, the results I ran out are PSNR =24.96 and SSIM =0.9266, which are quite different from the results in your paper.Set the following
Learning_rate: 0.001
crop_size: [240, 240]
train_batch_size: 24
val_batch_size: 1
network_height: 3
network_width: 6
num_dense_layer: 4
growth_rate: 16
Lambda_loss: 0.04
category: indoor
1399 images in ITS were selected from the indoor training data set

proteus1991 commented

Hi, sorry for the late reply. I'm very busy recently. Have you solved this problem by retraining from scratch?

shenhongqian-123456 commented
gfgc522 commented

@shenhongqian-123456 Hello, I have the same problem as you, can you tell me how to solve it？Thanks.

Deleted user commented

@shenhongqian-123456 Would you please share with us how you solved it? thanks

Deleted user commented

@gfgc522 Hi, were you able to solve it? thanks!

shenhongqian-123456 commented