honeygupta / uw-net Goto Github PK

Hi,
Does the SI-MSE refer actually to MSE or RMSE?

Thank you

Thanks for your reply, but how did you prepare the ground-truth images. Since in evaluate_metrics.py line 41, the gt between 0-255 will all be converted to 0.

Originally posted by @jinjidelinmouren in #10 (comment)

Hi,

I cannot find the code of your reference paper:

paper 5, writen by Drews and
paper 7, writen by peng.

How to finetune pre-trained model on new datasets? I have some new underwater images, I want to ask how to use the new RGB images fine-tuning pre-trained model. Do I need to use RGBD data when training?
If possible, would you mind provide the D-Hazy data set in the form of .npy files？

Hi!
I felt confused about the weights for each loss,
in your paper:
loss = 5Lgan + Lssim + Lgrad + Lcyc
however in the code, your loss is different:
loss = Lgan + 5Lcyc + 0.25Lssim(each) + Lgrad(only one for fakeA)

I'm coding on pytorch, and don't know how to set the weights for each loss.

Best regards,
Dong

Hi, thank you for the code!
I wanted to ask you something about the dataset preparation. I have some terrestrial RGB images and their depth. I compose the RGBD images (about 30) and then I have 100 underwater images.
I'm getting NaN for all the losses and it doesn't change with the time.

I wanted to ask so do I need to pair the images in the csv file?
In your case, how much time it needed to converge?

Thank you in advance

Michele

Sorry to bother. I try to convert it with its evaluation code，but I don't know its max depth when I save it. Thanks.

hey, would it be possible for you to adjust the code so that I don't need .npy files for testing? Or could you explain how to obtain the .npy files (for testing)

Thank you for the pre-trained model and test code.
I'm a little confused about evaluating the Berman dateset. Could you please share the evaluation code for Berman you showed in Tab 1?