train with multiple GPUs

[AGI-player]

Hi,

I get some problems in self-training process in st3d++ branch:

1)Perhaps the "consistency_matching" in some yamls should be "consistency_ensemble"

2)When using "tools\cfgs\da-nuscenes-kitti_models\secondiou_st3d\secondiou_st3d++_ros_car.yaml" by running "python train.py --cfg_file xxx" is ok, but I get error "can not find pesudo label for frame: xxx" by running “sh scripts/dist_train.sh xxx --cfg_file xxx”. The self-training process with multiple GPUs didn't work.

I wonder if someone get the same problems.

Thanks.

[jihanyang]

(1) Yes! This is a mistake caused by different name in internal and released version, I will fix it soon. Thank you so much!
(2) Check that whether this issue #9 can help?

[AGI-player]

Thanks for your reply! It works!