Replicating DSRGAN.ipynb and using other planes (Trans-ventricular, Trans-thalamic, and Trans-cerebellum)

[mxochicale]

Qingyu Yang reported the following error

Also I noted that notebook is not updated !pip install -qqq medisynth https://github.com/budai4medtech/xfetus/blob/main/examples/difussion-super-resolution-gan/DSRGAN.ipynb

[mxochicale]

sorted out

but leave it open in case there is further feedback from Qingyu Yang

[Qingyu-Yang1]

When calculating the FID, do I need to calculate the FID for the GAN network or for the diffusion model, or do I need both?

[mxochicale]

FID score is computed with the following lines from this notebook: https://github.com/budai4medtech/xfetus/blob/main/examples/difussion-super-resolution-gan/DSRGAN.ipynb (also available in google colabs https://colab.research.google.com/drive/1Cbudr2g5qdC2LGBj_xYS-amgJQ-6OVM6?usp=sharing). I would suggest to running all the notebook in your google colabs and then read sr_gan_loss.csv and plot FID values. Please share your google-colab to have a look to your development.

It would be great if you get FID for both GAN and diffusion model.

Let me know how it goes. Thanks, --Miguel

        # Calculate FID score using unaugmented images and fake images
        fake_images = torch.from_numpy(fake_images)
        fake_images = fake_images.to(device)
        fid.update(original_images.byte(), real=True)        
        fid.update(fake_images.byte(), real=False)
        current_fid = fid.compute().item()


        # Save model weights
        torch.save(netG.state_dict(), "SRGAN_G_x256" + str(epoch))
              
    # Write loss/FID to a log file for each epoch
    with open('sr_gan_loss.csv', 'a') as f_object:
        writer_object = writer(f_object)
        writer_object.writerow([str(epoch),  str(total_g_loss / 236), str(total_d_loss / 236), str(current_fid)])
        f_object.close()
    fid.reset()

[Qingyu-Yang1]

When I run the notebook, only one value appears for FID, and only three FID values appear after running three labels. In this case, how can I draw FID value to achieve that image?

It is appreciated for answering my question in your busy time.

--Qingyu Yang

[mxochicale]

Hi Qingyu, you need to read sr_gan_loss.csv to then plot it with matplotlib. This link might be useful to create your plots https://www.tutorialspoint.com/plot-data-from-csv-file-with-matplotlib.