search

MaybeShewill-CV / attentive-gan-derainnet

Unofficial tensorflow implemention of "Attentive Generative Adversarial Network for Raindrop Removal from A Single Image (CVPR 2018) " model https://maybeshewill-cv.github.io/attentive-gan-derainnet/
MIT License
262 stars 83 forks source link

Question #88

Closed Kamalhsn closed 2 years ago

Kamalhsn commented 2 years ago

Hi, Your work is so impressive. At present, I am working on a real dataset where I have to remove raindrops. I found that you prepared a mask image with _diff_image >= 35 (Ref: attentive-gan-derainnet/data_provider/tf_io_pipline_tools.py/line-92). However, I am not getting why did you do that particularly. Can I get more information about this mask image preparation? Thanks in advance.

MaybeShewill-CV commented 2 years ago

@Kamalhsn That's a threshold which you may adjust to suit your local dataset:)

Kamalhsn commented 2 years ago

OK. Can you please provide the criteria to choose the threshold value to extract all the raindrops? I mean, any resources?

Kamalhsn commented 2 years ago

Hi, How did you choose number of iterations to generate attention maps and didn't you include the attention loss in the model training?

Jack-Crum commented 2 years ago

I don't have the file in the path of ./data/vgg16.npy Therefore, I have no idea to run your train_model.py Can anyone help to upload that file? Thank you a lot! My eami address is 2629946925@qq.com

Jack-Crum commented 2 years ago

Can you help me to upload the vgg16.npy file? Thank you a lot!

MaybeShewill-CV commented 2 years ago

@Jack-Crum 随便百度一下就有了啊==！ https://blog.csdn.net/lqp888888/article/details/80699125

MaybeShewill-CV commented 2 years ago

@Kamalhsn parameters were based on the origin paper. You may find the details from the origin paper:)