box size

[dhiraj82]

Hi
It's not really an issue but I have a box size of 324 which is not divisible by 8. even if I am downsampling it by 2 or 3 or 4, it will not be divisible by 8. How can I get around this problem? is it ok to downsample it to 128? then how am i going to train cryodrgn without downsampling?

Thank you
Dhiraj

[ryanfeathers]

Hi Dhiraj,

You can definitely downsample by other factors! When downsampling to 128, multiply your pixel size by 324/128 and provide this value for cryodrgn analyze --Apix

Best,

Ryan

[michal-g]

As of v3.3.1 we are removing the error that is thrown when train_vae and train_nn inputs are not divisible by 8, instead throwing a warning that AMP training is not optimized when not using --no-amp (divisibility by 8 is only an issue when using AMP).

But yes, it is ok to downsample to 128 as Ryan answered above — you may not be able to run reconstruction as fast as otherwise, but you should't encounter any other issues because of this.