hellopipu
commented
8 months ago The released models for the CMRxRecon dataset have been trained with acceleration factors of 4, 8, and 10.
The model currently released for the multi-coil knee dataset was trained and tested solely with an acceleration factor of 8 to allow for a fair comparison with HUMUS-Net, which provides pre-trained model only at this acceleration factor. We intend to release the model trained with acceleration factors of 4 and 8 soon.
In general, our approach should be to train a single model on all available acceleration factors for the given dataset.
Hello,
In the context of both cardiac and knee challenges, there are multiple acceleration rates available. For cardiac scans, there are 4, 8, and 10-fold accelerated scans, and for knee scans, there are 4 and 8-fold accelerated scans. I'm seeking clarification on the training and testing procedure for the network:
Was the network trained exclusively with 4-fold accelerated scans and then tested with 4, 8, and 10-fold accelerated scans?
Or, was the network trained with 4-fold accelerated scans and tested only with 4-fold accelerated test scans?
Additionally, I'd like to know if the same approach extends to other acceleration rates.
Thank you,