about batch_size
Dear author: in your paper, for 3DSSD, you said that " The total batch size is set to 16, equally distributed on four NVIDIA V100 GPUs". But in your code file: 3dssd_sasa.yaml, the batch size is set to 8. I wonder that can i reproducing your results just follow the original 3dssd_sasa.yaml without any change?
Hi,
We didn't observe large performance difference between dividing the total batch size of 16 into 8x2 and 4x4.
If you would like to use the 4x4 division, you could simply modify the BATCH_SIZE_PER_GPU to 4 in 3dssd_sasa.yaml.
Thanks very much!!! I understand, if i only have one gpu, then i just need to directly set the total batch size to 16, am i right?