RoboBEV Benchmark

August 22, 2023 ยท View on GitHub

RoboBEV Benchmark

TPVFormer

CorruptionmIoU ptsmIoU vox
Clean26.8452.06
Cam Crash13.6527.39
Frame Lost11.9122.85
Color Quant21.4938.16
Motion Blur20.7938.64
Brightness25.7649.00
Low Light19.7837.38
Fog24.7846.69
Snow9.4919.39

Experiment Log

Camera Crash

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 18.76 | 37.49 | | Moderate | 9.19 | 19.88 | | Hard | 13.01 | 24.82 | | Average | 13.65 | 27.39 |

Frame Lost

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 19.55 | 37.53 | | Moderate | 10.51 | 20.13 | | Hard | 5.69 | 10.89 | | Average | 11.91 | 22.85 |

Color Quant

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 26.30 | 49.76 | | Moderate | 23.36 | 40.73 | | Hard | 14.83 | 23.99 | | Average | 21.49 | 38.16 |

Motion Blur

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 25.70 | 48.04 | | Moderate | 20.03 | 36.77 | | Hard | 16.66 | 31.13 | | Average | 20.79 | 38.64 |

Brightness

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 26.48 | 51.17 | | Moderate | 25.89 | 49.34 | | Hard | 24.91 | 46.51 | | Average | 25.76 | 49.00 |

Low Light

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 23.44 | 44.20 | | Moderate | 20.39 | 38.31 | | Hard | 15.53 | 29.65 | | Average | 19.78 | 37.38 |

Fog

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 25.30 | 48.04 | | Moderate | 24.92 | 47.10 | | Hard | 24.12 | 44.95 | | Average | 24.78 | 46.69 |

Snow

| Severity | mIoU pts | mIoU vox | | :----------: | :-----: | | | | Easy | 14.50 | 28.52 | | Moderate | 7.34 | 15.51 | | Hard | 6.64 | 14.16 | | Average | 9.49 | 19.39 |

References

@inproceedings{huang2023tri,
  title={Tri-perspective view for vision-based 3d semantic occupancy prediction},
  author={Huang, Yuanhui and Zheng, Wenzhao and Zhang, Yunpeng and Zhou, Jie and Lu, Jiwen},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={9223--9232},
  year={2023}
}