GPU nums and and training time

[Debrove]

Hi! It's a nice job!

It seems use data augmentation for nuscenes datasets, maintaining 7724 samples (total batch sizes=16), which is more than the original number of nuscenes samples.

I want to know the gpu nums you used and how long to pretrain the swin-base Transformer.

Looking forward to your reply! Thank you!

[jihaonew]

Hi!

We only use the train set for pretraining. Can you describe in more detail how you obtained "7724 samples"?

The base model is pretrained with 16 gpus for about 2 days ( 50 epochs).