search

ZAX130 / RDP

23 stars 0 forks source link

datasets problem #1

Open LinjieFu-U opened 8 months ago

LinjieFu-U commented 8 months ago

Hello, can you provide the processed datasets

ZAX130 commented 8 months ago

I'm so sorry that I'm afraid I can't, because the size of the total dataset is too big. However, we have further published the link of the IXI dataset processed by freesurfer and an example code of how to create a pkl dataset from the original dataset, which may help with this issue.

ZAX130 commented 8 months ago

We have published the pkl-format LPBA dataset we used in experiments.

LinjieFu-U commented 8 months ago

Thank you very much for your contributions. We were able to reproduce results similar to those in your work on both IXI and LPBA datasets, but when we tried to use the data in the link of Mindboggle dataset you provided for the experiment, the DICE of the experiment result was only 0.49, far lower than the result of 0.62 in your work. We suspect that it is a problem when we process the data. Could you please provide the processed Mindboggle dataset?

ZAX130 commented 8 months ago

Have you ever tried to change the Seg_norm() in data/trans.py and dice_val_VOI in utils.py?

LinjieFu-U commented 8 months ago

Yes, we used the seg_norm() and dice_val Settings for the OASIS dataset in Transmorph

ZAX130 commented 8 months ago

But you need to change them according to the labels in Mindboggle. You can print all the classes in the label data.

LinjieFu-U commented 7 months ago

Sorry to bother you, but I have another question. How should I calculate the dice of each region separately as in the paper.

ZAX130 commented 7 months ago

Please check the code in FAIM: https://github.com/dykuang/Medical-image-registration/blob/master/source/view_dice.py