---
license: apache-2.0
---

# FlowOcc3D (3D Occypancy Flow Dataset)

<img src="./FlowOcc3D.jpg" width="800"/>

Download [FlowOcc3D](https://huggingface.co/viewformer/ViewFormer-Occ/blob/main/occ_flow_sparse_ext.zip). Unzip it in `./data/nuscenes`.

Our FlowOcc3D assigns a flow vector to each foreground occupancy of [Occ3D](https://github.com/CVPR2023-3D-Occupancy-Prediction/CVPR2023-3D-Occupancy-Prediction).

Here we briefly introduce how to use FlowOcc3D. We store the flow and index of each foreground voxel in the `xxx.bin` file and `xxx_idx.bin` file.
```python
W, H, Z = 200, 200, 16
sample_idx = results['sample_idx'] # nuScenes sample token
data_path = os.path.join('./data/nuscenes', 'occ_flow_sparse_ext', sample_idx)

occ_flow = np.ones((W*H*Z, 2)) * pad_value # pad_value could be zero
sparse_flow = np.fromfile(data_path + '.bin', dtype=np.float16).reshape(-1, 3)[:, :2]
sparse_idx = np.fromfile(data_path + '_idx.bin', dtype=np.int32).reshape(-1)
occ_flow[sparse_idx] = sparse_flow
occ_flow = occ_flow.reshape(W, H, Z, 2)
```


## Citation

```bibtex
    @article{li2024viewformer,
        title={ViewFormer: Exploring Spatiotemporal Modeling for Multi-View 3D Occupancy Perception via View-Guided Transformers}, 
        author={Jinke Li and Xiao He and Chonghua Zhou and Xiaoqiang Cheng and Yang Wen and Dan Zhang},
        journal={arXiv preprint arXiv:2405.04299},
        year={2024},
    }
```