What does Batchsize epsilon mean?

3D Refinement
1

Hi, all

I’m wondering what actually this parameter means? :joy:

2

Hi @pywt901!

Batchsize epsilon controls the estimated proportion of Fourier pixels which will be missed by the minibatch. It is therefore a means of setting the minibatch size by specifying an average inaccuracy of the batch’s volume. Setting this value higher therefore allows for fewer particles in the minibatch, while a lower value creates minibatches with more particles. Note that this parameter should always remain above 0.

That being said, if you want to adjust the batch size, it’s almost always better to use Batchsize snrfactor, which straightforwardly multiplies the batchsize (i.e., doubling Batchsize snrfactor doubles the minibatch).