GPU and training
Yurushia1998 opened this issue · comments
I'm testing DetectoRS training using only 1 GPU with 1 image per gpu, 1 worker per gpu and the cuda memory is full already. My GPU has 11GB in memory. How comes a detection models can take this much memory, or is there a chance that I do anything wrong? What is the requirement Cuda memory per gpu for this models?
@Yurushia1998 have you solve this problem? i meet the same, the gpu memory is too high~