search

wongzbb / DiffMa-Diffusion-Mamba

Soft Masked Mamba Diffusion Model for CT to MRI Conversion (Official PyTorch Implementation)
MIT License
47 stars 3 forks source link

some problems about result using sample.py #7

Open rainchan0227 opened 2 months ago

rainchan0227 commented 2 months ago

Hi, Congratulations to your great work, and thanks for making open-source! I used the "DiffMa-B/2" and the "DiffMa/1000000.pt", I used some samples from the dataset you provided. However, the results of images generated is quite poor. I wonder if this has some details need to changed. Looking forward to your reply！ (Below are CT, image generated and original MRI) 16_sample_ct 16_sample_gen 16_sample_ori

wongzbb commented 2 months ago

Hi, you can adjust the window width and window level during training. Try setting those parameters and re-training—it might work better!

rainchan0227 commented 2 months ago

Hi, you can adjust the window width and window level during training. Try setting those parameters and re-training—it might work better!

Thanks for your suggestion, I am now using the dataset you provided for model training, this is the result after 0_150_000 iterations, you can see that the results have changed somewhat, hopefully the quality of reconstruction can be improved by continuing the training later. If possible could you provide me with the brain model parameter files used in the article results and the brain CT-MRI samples used (one slice will do), thank you very much!

4_sample_ct 4_sample_gen 4_sample_ori

rainchan0227 commented 2 months ago

Also I would like to ask if the mask data is automatically generated based on CT or manually labelled, and if it is automatically generated by what means please, thank you very much!

zfw-cv commented 2 months ago

Also I would like to ask if the mask data is automatically generated based on CT or manually labelled, and if it is automatically generated by what means please, thank you very much!

Hello, friend. I came across your question by chance. I checked the synthRAD datasets article and found this description. “The binary mask was generated using a thresholding technique and hole-filling algorithms from the ITK image processing toolkit.” I hope it helps. Of course, I also look forward to the author’s confirmation. Thank you.

rainchan0227 commented 2 months ago

thanks a lot!

TaoZhong11 commented 2 days ago

Hi friends, I got the same problems as yours when I sampled the cases by using the pre-trained DiffMa model. So did you solve the problems when you retrain the models by yourself?

Hi, Congratulations to your great work, and thanks for making open-source! I used the "DiffMa-B/2" and the "DiffMa/1000000.pt", I used some samples from the dataset you provided. However, the results of images generated is quite poor. I wonder if this has some details need to changed. Looking forward to your reply！ (Below are CT, image generated and original MRI) 16_sample_ct 16_sample_gen 16_sample_ori