rui1996/deraindrop

Attentive Generative Adversarial Network for Raindrop Removal from A Single Image (CVPR 2018)

deraindrop's Issues

Which part of the code is used in ConvLSTM, and can you provide some training code?

Can this demo used for real time video ?

I was reading about this project and I noticed it's only for a single image. so is it possible this can be done in real time?
Thank you for your help.

How to Get the mask M?

Hello! Thank you for releasing the code. I don't know how you get the binary mask M in the following formula. Could you tell me how you did it？Thank you!

can you release some training details?

Hey Rui,

Your work is interesting and amazing!!!
Could you please release some training details like batch size, input size?
I guess you input the whole image into the net without down-sampling or cropping.

Best,
Zewei

HI, do you have the dataset download link which can be used in the mainland?

HI, do you have the dataset download link which can be used in the mainland? such as Baidu yun?
I try many times , but I can not get the file from the google download link " https://drive.google.com/open?id=1e7R76s6vwUJxILOcAsthgDLPSnOrQ49K"

Confusion of the loss function

hi!
I wonder whether the first term of Eq.(9) will become larger, when the output O of Generator is closed to target after a lot of epochs , which means it is hard for Discriminator to infer attention map from O. So, does the loss function in Eq.(9) work after several epochs?

How to Get the Ground Truth?

Hello! Thank you for releasing the code. I don't know how you get the ground truth of a picture with raindrop. Could you tell me how you did it？Thank you!

Could you please share the training dataset

If that's all right could you please share your training dataset. Thanks for your kindness

How to run the demo with my own dataset?

Can you please explain how to run this demo using my own dataset?
Thank you

How to train the model and which kind of GPU for training?

Thanks for sharing this amazing code, but I have a question that how to train the model and which kind of GPU for training? I only find predict.py, could you mind providing train.py for me?

release more information about the training details

Could you please release more information about the training details such as batch_size, learning rate , epoch nums if it matters nothing? Thanks a lot

test my own graphics without any effect

Excuse me, why do I test my own graphics without any effect?In addition,does the generated graph of the test have no attention map?

请问用的什么GPU

谢谢作者的工作，我用1080的GPU运行demo都超出内存了，请问用的什么GPU训练和测试呢？谢谢

Can you provide the code for training? Thanks

Can I know how you have generated the pkl file in binary format?I have tried it but I am able to create in text format and unable to create it in binary format.Please help me with this issue..

> 我是我根据论文自己复现的基于pytorch的代码 https://github.com/7568/DeRaindrop ， 里面有训练代码和测试代码

我是我根据论文自己复现的基于pytorch的代码 https://github.com/7568/DeRaindrop ， 里面有训练代码和测试代码

@7568 你好，我用你的代码进行训练，训练结果和原文给出的权重差一点。这和您没有使用dataloader、和没有使用动态学习率有关吗

Originally posted by @zrt791521360 in #10 (comment)

Differences

Hi I want to tell you 2 thing to update.
1- You need to update your readme.md file. Your datasets are invalid now. So you need to change or add a new orientation like:

• Change the value of argument "baseroot" in file "./test.sh" to the path of testing data. Be carefull when yo loading your own images. The image loader looks at baseroot/small/rain and baseroot/small/unrain. Corresponding images under this files need to be have same name!

This is for 3'rd point of testing.

2- You need to add these to the for loop at 87. line, validation.py because of the difference of datatype of height_origin and width_origin;

        height_origin = int(height_origin.item())
        width_origin = int(width_origin.item())

mistake in discriminator.py？

the code in discriminator.py（68-70）：
“”
x = self.conv7(x * mask)
x = self.conv7(x)
x = self.conv8(x)
“”
is "x = self.conv7(x)" extra？you can't run the code with “x = self.conv7(x)” in it

the discriminator is not the discriminator described in your CVPR paper.

Is there something wrong???

Or on purpose