search

lcbkmm / TC-KANRecon

TC-KANRecon: High-Quality and Accelerated MRI Reconstruction through Adaptive KAN Mechanisms and Intelligent Feature Scaling
17 stars 1 forks source link

training problem (data-loader) #1

Open tom-kh opened 1 week ago

tom-kh commented 1 week ago

Thank you for your hard work. I am facing issues with specifying the dataset path and need further guidance to proceed. I want to replicate your code, but I'm unable to move forward due to the path issue. I've downloaded the SKM-TEA dataset, but I'm unclear about the additional steps required to feed it into your network for training from scratch. Also, since no pre-trained weights are available, could you please provide some guidance on how to proceed? Your assistance would be greatly appreciated.

lcbkmm commented 3 days ago

Certainly, thank you very much for your recognition and support. I'm sorry to hear that you encountered some difficulties while setting up the dataset path. To help you better integrate the SKM-TEA dataset into your code and smoothly start the training process, I'd like to provide you with some detailed steps and suggestions that I hope will be helpful to you.

First, please ensure that you have correctly downloaded and unzipped the SKM-TEA dataset. It's worth mentioning that in our experiments, we used single-coil data, so please confirm that the data you are using is also the single-coil data we require to avoid any issues during subsequent processing. Secondly, the SKM-TEA dataset is downloaded in h5 format, but for the convenience of subsequent processing and training, we need to convert it to a pt file format that contains k-space information. After converting the file format, you need to divide the obtained pt file data into two files: training dataset and testing dataset. This process can be divided according to your actual needs, but to ensure the accuracy of training and testing results, it is recommended that you maintain data balance and representativeness when dividing. Finally, when entering your data file path in the configuration file, please make sure the path is correct and complete. If you're unsure about the path, you can perform a simple test to ensure the data can be loaded and read correctly. Thank you again for your recognition and support. I wish you great success in using the SKM-TEA dataset!