FengheTan9
Issues when running the code
Hi, dear authors, when runnning your code to perform SSL, it encounters an error which seems to be caused by the missing definition of MambaSparK which appears in line 157 'MambaSparK.forward' is this a typo?
By the way, may i ask you how many subvolumes do your methods use and how long it take for the SSL stage. I pretrain the methods with 125k subvolumes, with the shape of 19219216, it seems the method will spend a very long time to perform ssl....
Hi, dear authors, when runnning your code to perform SSL, it encounters an error which seems to be caused by the missing definition of MambaSparK which appears in line 157 'MambaSparK.forward' is this a typo?
fixed !
By the way, may i ask you how many subvolumes do your methods use and how long it take for the SSL stage. I pretrain the methods with 125k subvolumes, with the shape of 192_192_16, it seems the method will spend a very long time to perform ssl....
Hello, the computational bottleneck of MambaMIM currently lies in TOKI, which is indeed quite slow at the moment. 🥵 For reference, training on a 6.8K volume takes about one hour per epoch. For faster training, we suggest that you consider fine-tuning the model using the pretrained weights we have made available. 🤗🤗🤗