Batch-norm with Variable batch size (where sometimes the batch can be of size 1 too))

[zimmer550 (Sarim Mehdi)]

So, my question is basically a follow-p of this:

Variable batch size during training

Hi. I am training an RNN encoder-decoder to make predictions of the trajectory of agents. I am using the KITTI dataset. The training dataset consists of 21 image folders and each folder has a variable number of images. So, this means that the batch size changes during training, for example: A batch size of one, when training an encoder, means one sequence. Keeping that in mind, let’s say I take as input 5 images and, while decoding, also works on 5 images. So, that is a total of 10 images. So, …

Is batch-norm suitable for my situation where the batch size can fluctuate between 1 and 10? I know that there is no batchnorm for batch size of 1. So, is that a hard no for my situation or is there something I could do?

[SimonW (Simon Wang)]

with that low batch size, you are probably better just not use batch norm at all.