Please ensure you have prepared the environment and the dataset.

Train and Test

Train

1. Train PreWorld with 8 GPUs on nuScenes dataset (3D Occupancy Prediction):

   bash ./tools/dist_train.sh ./configs/preworld/nuscenes/preworld-7frame-finetune.py 8
   

   Or:

   # Pre-training Stage
   bash ./tools/dist_train.sh ./configs/preworld/nuscenes/preworld-7frame-pretrain.py 8
   # Fine-tuning Stage
   bash ./tools/dist_train.sh ./configs/preworld/nuscenes/preworld-7frame-finetune.py 8 --resume-from=./path/to/pre-trained_ckpts.pth  # Please modify 'max_epochs' in the configuration file accordingly.
   

2. Train PreWorld with 8 GPUs on nuScenes dataset (4D Occupancy Forecasting):

   bash ./tools/dist_train.sh configs/preworld/nuscenes-temporal/preworld-7frame-finetune-traj.py 8
   

   Or:

   # Pre-training Stage
   bash ./tools/dist_train.sh ./configs/preworld/nuscenes-temporal/preworld-7frame-pretrain-traj.py 8
   # Fine-tuning Stage
   bash ./tools/dist_train.sh ./configs/preworld/nuscenes-temporal/preworld-7frame-finetune-traj.py 8 --resume-from=./path/to/pre-trained_ckpts.pth  # Please modify 'max_epochs' in the configuration file accordingly.
   

Evaluation

Eval PreWorld with 8 GPUs on nuScenes dataset (3D Occupancy Prediction):

bash ./tools/dist_test.sh ./configs/preworld/nuscenes/preworld-7frame-finetune.py ./path/to/ckpts.pth 8

Eval PreWorld with 8 GPUs on nuScenes dataset (4D Occupancy Forecasting):

bash ./tools/dist_test_temporal.sh ./configs/preworld/nuscenes-temporal/preworld-7frame-finetune-traj.py ./path/to/ckpts.pth 8

Visualization

3D Occupancy Prediction on nuScenes dataset:

# Dump predictions
bash ./tools/dist_test.sh ./configs/preworld/nuscenes/preworld-7frame-finetune.py ./path/to/ckpts.pth 1 --dump_dir=vis_dirs/output
# Visualization (select scene-id)
python tools/visualization/visual.py vis_dirs/output/scene-xxxx

(The pkl file needs to be regenerated for visualization.)