Error on pre-trained model loading on nuScenes dataset about openscene HOT 1 CLOSED

Hi, I just noticed this error was caused by When loading a pre-trained model from a checkpoint, the layer names might have prefixes like "module." due to the use of torch.nn.DataParallel during training.

So I added two lines before pre-trained model loading,
state_dict =checkpoint['state_dict']
state_dict = {k.partition('module.')[2]:state_dict[k] for k in state_dict.keys()}

Now the command "sh run/eval.sh out/nuScenes_openseg config/nuscenes/ours_openseg_pretrained.yaml ensemble" can work.

I just wondering if there are any more elegant ways to solve this problem. Like setting "use_shm=True" to make pre-trained model testing support multi-GPU?

from openscene.