Open yxchng opened 4 days ago
the gpu util of this code seem to be very low, do you have any idea of where the bottleneck is and how to optimize?
hard to say anything specific. may be try to increase num_workers in dataloader, as we use some slow augmentations like ElasticTransfrom
num_workers
ElasticTransfrom
the gpu util of this code seem to be very low, do you have any idea of where the bottleneck is and how to optimize?