search

vigsivan / RWCNet

Official implementation of Recurrence with Correlation Network for Medical Image Registration
MIT License
8 stars 4 forks source link

config file and TRE metric not calculating in evaluation phase for NLST dataset. #5

Open animesh-007 opened 1 year ago

animesh-007 commented 1 year ago

Hi @vigsivan, I am using the given train_config.json for training the model on the NLST dataset. I also used the eval.py to evaluate the trained model. But in the evaluation phase, eval.py only calculates MSE and not TRE metric.

In the train_config.json, values of dset_min: -3024 and dset_max: 15622 will be these right for the NLST dataset? I have calculated these values using get_dset_minmax.py. Will there be any other change in the config file for the NLST dataset?

Can you please check the eval.py for the NLST dataset evaluation?

vigsivan commented 1 year ago

I believe the min, max we use for NLST is -4000:4000, but let me get back to you later today.

Re: eval, are you running into issues?

animesh-007 commented 1 year ago

Yes, in the current version of eval.py file it expects fixed_keypoints and moving_keypoints, which are not present in the data.jon file. Because of which it is not calculating TRE metric for the NLST dataset.

Also, can you share the the min, and maximum value for different datasets used by you?

animesh-007 commented 1 year ago

Hi, @vigsivan Did you get a chance to look at eval.py for the NLST dataset? Can you please check once if it is working fine?

animesh-007 commented 1 year ago

Hi @vigsivan. Hope you are doing good. Can you please check how we can calculate the TRE metric with your evaluation code?