1. Download nuScenes V1.0 full dataset data HERE. Folder structure:
AdaptiveOcc
├── data/
│ ├── nuscenes/
│ │ ├── maps/
│ │ ├── samples/
│ │ ├── sweeps/
│ │ ├── v1.0-test/
│ │ ├── v1.0-trainval/
2. Download the generated train/val pickle files and put them in data.
3. Download the generated dense occupancy labels from SurroundOcc (resolution 200x200x16 with voxel size 0.5m) and put and unzip it in data.
| resolution | Subset | Link | Size |
|---|---|---|---|
| 200x200x16 | train | link | 3.2G |
| 200x200x16 | val | link | 627M |
Please note that:
- the shape of each npy file is (n,4), where n is the number of non-empty occupancies. Four dimensions represent xyz and semantic label respectively.
- In the dataloader, it convert empty occupancies as label 0 and ignore class as label 255.
- The occupancy labels are the voxel indexes under LiDAR coordinate system, not the ego coordinate system. You can use the code to convert voxel indexes to the LiDAR points.
Folder structure:
SurroundOcc
├── data/
│ ├── nuscenes/
│ ├── nuscenes_occ/
│ ├── nuscenes_infos_train.pkl
│ ├── nuscenes_infos_val.pkl