mahfuzmohammad/healthygan

Official PyTorch implementation of HealthyGAN - SASHIMI 2022

healthygan's Issues

discriminator much more powerful

Dear @mahfuzmohammad,

I notice from the codes that the discriminator is set to be stronger than the generator (the discriminator has way more parameters). May I know the reason for that? Thank you!

Environment issues when running

Hello,

While running in server environment with custom data,

Could not load library libcudnn_cnn_infer.so.8. Error: libcuda.so: cannot open shared object file: No such file or directory
Please make sure libcudnn_cnn_infer.so.8 is in your library path!

I'm trying to run it in a Docker environment because I get an error like,
If you have any docker files used, could you please help?

difficulty in undertsanding the paper

Hi @mahfuzmohammad:
Thank you for sharing the fantastic work! I am an undergraduate student studying utilizing GAN for anomaly detection. While reading your paper, I find it difficult to follow the idea of using attention masks. Would you mind sharing some literature to help me understand this idea? Thank you so much!

Learning and utilizing datasets from other tasks

Hello,

I would like to apply your architecture to other medical domain tasks. (ex : CT, MRI ... )
Have you ever tried this and, if so, was the result good?

I look forward to your reply, thank you. :)

Asking about Covid dataset

I follow the link about dataset that you provide. It seem they use data from Kaggle (https://www.kaggle.com/datasets/andyczhao/covidx-cxr2?select=competition_test). However, when I download and extract to the path based on your instruction. I found that the data from Kaggle are not fit with your pickle file. Would you please suggest me how to generate the covid dataset that you use.

understanding the paper

Dear @mahfuzmohammad ,

Thanks so much for your prompt replies to my previous issues! I still have some difficulty understanding the idea. May I know how to explain this sentence:

To be specific, if the input image is a healthy image, the generator is expected to behave like an autoencoder. If the input is a diseased image, the generator should remove anomalous parts and produce a healthy image in the output.

Since the model itself doesn't know its input here (which comes from the mixed dataset), how is the model supposed to do so?

Also from my own experiments, it seems that the masks are outputting only 1s (in the plots, they look completely white) for anomalies, so essentially the mask might not only activate on anomaly regions. However, the method is still effective at detecting anomalies. May I know how to explain this effectiveness?

I greatly appreciate your response!

Asking about code

I am trying to reproduce the project and having difficulties understanding the code. Let this issue be the thread where I am asking anything about the project.

First question.
I see this code to produce outputs of Discriminator model:

    def forward(self, x):
        h = self.main(x)
        out_src = self.conv1(h)
        return h, out_src

I see that out_src is used in training, but I am not finding h used anywhere. What is h here? And how is it used in the project?

Thank you.