RoboBEV Benchmark

SurroundOcc

Corruption	IoU	mIoU
Clean	0.3149	0.2030
Cam Crash	0.1996	0.1160
Frame Lost	0.1810	0.1000
Color Quant	0.2584	0.1403
Motion Blur	0.2257	0.1241
Brightness	0.3072	0.1918
Low Light	0.2479	0.1215
Fog	0.2964	0.1842
Snow	0.1834	0.0739

Experiment Log

Camera Crash

Severity	IoU	mIoU
Easy	0.2358	0.1421
Moderate	0.1747	0.1015
Hard	0.1884	0.1044
Average	0.1996	0.1160

Frame Lost

Severity	IoU	mIoU
Easy	0.2522	0.1549
Moderate	0.1686	0.0907
Hard	0.1223	0.0545
Average	0.1810	0.1000

Color Quant

Severity	IoU	mIoU
Easy	0.3057	0.1906
Moderate	0.2679	0.1491
Hard	0.2016	0.0814
Average	0.2584	0.1403

Motion Blur

Severity	IoU	mIoU
Easy	0.2854	0.1797
Moderate	0.2107	0.1093
Hard	0.1810	0.0834
Average	0.2257	0.1241

Brightness

Severity	IoU	mIoU
Easy	0.3133	0.2008
Moderate	0.3081	0.1925
Hard	0.3004	0.1822
Average	0.3072	0.1918

Low Light

Severity	IoU	mIoU
Easy	0.2734	0.1537
Moderate	0.2505	0.1249
Hard	0.2199	0.0861
Average	0.2479	0.1215

Fog

Severity	IoU	mIoU
Easy	0.3047	0.1909
Moderate	0.2988	0.1860
Hard	0.2858	0.1759
Average	0.2964	0.1842

Snow

Severity	IoU	mIoU
Easy	0.2082	0.1007
Moderate	0.1770	0.0660
Hard	0.1650	0.0551
Average	0.1834	0.0739

References

@article{wei2023surroundocc, 
      title={SurroundOcc: Multi-Camera 3D Occupancy Prediction for Autonomous Driving}, 
      author={Yi Wei and Linqing Zhao and Wenzhao Zheng and Zheng Zhu and Jie Zhou and Jiwen Lu},
      journal={arXiv preprint arXiv:2303.09551},
      year={2023}
}