How many data are used for self-supervised pretraining?

[SxJyJay]

Hi Junbo,
Thanks for your great work and open-sourced code. I have a problem of the amount of training data used in self-supervised pretraining. You mentioned in the paper that the whole training set of Waymo is used for pretraining. However, I notice the config file for SSL that the "load_interval" is set to 4, which means that only 25% training data are used during pretraining.

ProposalContrast/configs/waymo/voxelnet/ssl_pretrain_voxelnet.py

Line 113 in 3b02290

load_interval=4

[mu-cai]

I think so. In paper's claim, you should use load_interval=1.

[SxJyJay]

I think so. In paper's claim, you should use load_interval=1.

Thanks for your reply! By comparing the results I trained using load_interval=4 with those reported in the paper, it seems that using the whole waymo training set can bring 0.3~0.5 improvements in LEVEL_2 mAP and mAH across different classes.