Using dataparallel on microscopic images (20000x20000)

vision
1

Hi All,

I’m trying to develop an image segmentation deep learning pipeline for microscopic images at super-resolutions. Have two 2080 RTX Tis, 11gb vram each.

I obviously cannot get my batch size over 1, but unfortunately dataparallel only splits across the batch size. Is there an already available way to split my compute on a single image across different gpus with pytorch?

Thanks!

2

Hi,

I’m afraid these isn’t as it is not a very common use case.
You might be able to split the image yourself into a batch of smaller images and then reconstruct the result?