$bash run_docker.sh /data gpupack nuscenes dataset
# To pack train set
$python3 ./tools/nuscenes_packer.py --src-data-dir ./data/nuscenes/ --pack-type lmdb --target-data-dir ./tmp_data/nuscenes/v1.0-trainval --version v1.0-trainval --split-name train
# To pack val set
$python3 ./tools/nuscenes_packer.py --src-data-dir ./data/nuscenes/ --pack-type lmdb --target-data-dir ./tmp_data/nuscenes/v1.0-trainval --version v1.0-trainval --split-name valThe tree strcture of nuscenes packed is as following:
|-- tmp_data
| |-- nuscenes
| | |-- v1.0-trainval
| | | |-- train_lmdb #packed train set
| | | | |-- data.mdb
| | | | `-- lock.mdb
| | | `-- val_lmdb #packed val set
| | | | |-- data.mdb
| | | | `-- lock.mdb
Decompress v1.0-trainval_meta.tar and nuscenes-map-extension-v1.3.zip, then reconstruct as following:
|-- tmp_data
| |-- nuscenes
| | |-- meta
| | | |-- maps
| | | | |-- 36092f0b03a857c6a3403e25b4b7aab3.png
| | | | |-- ...
| | | | |-- 93406b464a165eaba6d9de76ca09f5da.png
| | | | |-- prediction
| | | | |-- basemap
| | | | |-- expansion
| | | |-- v1.0-trainval
| | | |-- attribute.json
| | | ...
| | | |-- visibility.json
| | `-- v1.0-trainval
| | | |-- train_lmdb
| | | `-- val_lmdb
modify configs according to your device
$cd /open_explorer/worksapce/
$python3 tools/train.py --config configs/bev_ipm_efficientnetb0_multitask_nuscenes.py --stage float$python3 tools/predic.py --config configs/bev_ipm_efficientnetb0_multitask_nuscenes.py --stage floatpython3 tools/infer.py --config configs/bev/bev_mt_ipm_efficientnetb0_nuscenes.py --model-inputs imagedir:${imagedir},homo:${homography.npy}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent