yuhuUSTC / DehazeDDPM

High-quality Image Dehazing with Diffusion Model
21 stars 1 forks source link

Inquiry Regarding First Stage Training and Test Results #5

Open Emily-29 opened 5 months ago

Emily-29 commented 5 months ago

Thank you for making your code accessible on GitHub. It has been invaluable for my project on image dehazing. I have a query regarding the initial phase of your model's training process that I hope you can clarify: could you elaborate on how the training is conducted during the first stage?

Furthermore, I attempted to apply the models "NH_net_g_80000.pth" and "NH_I230000_E4600_gen.pth" to image "55.png" from the NH-HAZE dataset. The outcomes did not align with my expectations. I would greatly appreciate it if you could review the attached results and advise if there might be any steps I have overlooked or misimplemented.

ret

yeziyuan27 commented 3 months ago

Hello! I have also encountered this problem. Have you solved this problem?

yuhuUSTC commented 3 months ago

The first stage follows the well-known ASM formulation: I(x) = J(x)t(x) + A (1 - t(x)). For the output in https://github.com/yuhuUSTC/DehazeDDPM/blob/5ee18db0646de74741714c83493a8d3f17c1a8c2/model/networkHelper.py#L676, out_J means the final output, stage1_output represents the output of the first stage, out_T denotes t(x), out_A denotes A, and out_I denotes I(x). Therefore, the training loss for the first stage is: loss = L1loss(HR, out_J) + L1loss(HR, stage1_output) + L1loss(LR, out_I).

yeziyuan27 commented 3 months ago

Thank you very much for your reply! I would like to know why the result of my validation with your pre-trained model on the 55-hazy images on the NH-haze dataset is a noised image?This question is the same as the second question raised by the host, looking forward to your reply! Thank you very much! 屏幕截图 2024-08-16 210107 屏幕截图 2024-08-16 210210 @yuhuUSTC

Emily-29 commented 3 months ago

Hello! I have also encountered this problem. Have you solved this problem?

Not yet

zhengchaobing commented 1 month ago

Hello! I have also encountered this problem. Have you solved this problem?

I have also encountered this problem too. Have you solved this problem now?

zhengchaobing commented 1 month ago

Thank you very much for your reply! I would like to know why the result of my validation with your pre-trained model on the 55-hazy images on the NH-haze dataset is a noised image?This question is the same as the second question raised by the host, looking forward to your reply! Thank you very much! 屏幕截图 2024-08-16 210107 屏幕截图 2024-08-16 210210 @yuhuUSTC

I have also encountered this problem too. Have you solved this problem now?