search

JeffWang987 / OpenOccupancy

[ICCV 2023] OpenOccupancy: A Large Scale Benchmark for Surrounding Semantic Occupancy Perception
Apache License 2.0
551 stars 49 forks source link

Batch size in training stage #35

Closed Jeffery-J closed 11 months ago

Jeffery-J commented 11 months ago

HI, I find that _samples_pergpu is set as 1 in all your config, so is there any reason to do so? I think the memory is enough in some tasks if you set _samples_pergpu to 2 or 4.

JeffWang987 commented 11 months ago

All models are trained with a batch size of 8 (one for each GPU) for fair comparison. And you can increase batch size during training.

Jeffery-J commented 11 months ago

Get it, thanks !

annopackage commented 11 months ago

Hi, how much memory does it need if using resnet50 and 6 cameras? If setting larger batch size, can we get better results?