RoboBEV Benchmark

The official nuScenes metrics are considered in our benchmark:

Average Precision (AP)

The average precision (AP) defines a match by thresholding the 2D center distance d on the ground plane instead of the intersection over union (IoU). This is done in order to decouple detection from object size and orientation but also because objects with small footprints, like pedestrians and bikes, if detected with a small translation error, give $0$IoU. We then calculate AP as the normalized area under the precision-recall curve for recall and precision over 10$10$$\mathbb{D}={0.5, 1, 2, 4}$meters and the set of classes$\mathbb{C}$ :

$$\text{mAP}= \frac{1}{|\mathbb{C}||\mathbb{D}|}\sum_{c\in\mathbb{C}}\sum_{d\in\mathbb{D}}\text{AP}_{c,d} .$$

True Positive (TP)

All TP metrics are calculated using $d=2$ m center distance during matching, and they are all designed to be positive scalars. Matching and scoring happen independently per class and each metric is the average of the cumulative mean at each achieved recall level above $10$$10$$1$.

• Average Translation Error (ATE) is the Euclidean center distance in 2D (units in meters).
• Average Scale Error (ASE) is the 3D intersection-over-union (IoU) after aligning orientation and translation ($1$ − IoU).
• Average Orientation Error (AOE) is the smallest yaw angle difference between prediction and ground truth (radians). All angles are measured on a full $360-degree period except for barriers where they are measured on a \180$-degree period.
• Average Velocity Error (AVE) is the absolute velocity error as the L2 norm of the velocity differences in 2D (m/s).
• Average Attribute Error (AAE) is defined as $1 minus attribute classification accuracy (\1$ − acc).

nuScenes Detection Score (NDS)

mAP with a threshold on IoU is perhaps the most popular metric for object detection. However, this metric can not capture all aspects of the nuScenes detection tasks, like velocity and attribute estimation. Further, it couples location, size, and orientation estimates. nuScenes proposed instead consolidating the different error types into a scalar score:

$$\text{NDS} = \frac{1}{10} [5\text{mAP}+\sum_{\text{mTP}\in\mathbb{TP}} (1-\min(1, \text{mTP}))] .$$

AutoAlignV2

Corruption	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Clean	0.6139	0.5649	0.3300	0.2699	0.4226	0.4644	0.1983
Cam Crash	0.5849	0.5137	0.3393	0.2710	0.4329	0.4786	0.1977
Frame Lost	0.5832	0.5104	0.3396	0.2718	0.4315	0.4787	0.1984
Color Quant	0.6006	0.5405	0.3347	0.2703	0.4254	0.4708	0.1950
Motion Blur	0.5901	0.5188	0.3336	0.2708	0.4243	0.4715	0.1923
Brightness	0.6076	0.5497	0.3310	0.2699	0.4231	0.4560	0.1922
Low Light	0.5770	0.4949	0.3390	0.2717	0.4157	0.4785	0.2000
Fog	-	-	-	-	-	-	-
Snow	-	-	-	-	-	-	-

Experiment Log

Camera Crash

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.5879	0.5200	0.3382	0.2702	0.4392	0.4745	0.1983
Moderate	0.5852	0.5144	0.3408	0.2710	0.4333	0.4778	0.1973
Hard	0.5816	0.5068	0.3390	0.2718	0.4262	0.4835	0.1976
Average	0.5849	0.5137	0.3393	0.2710	0.4329	0.4786	0.1977

Frame Lost

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.5965	0.5341	0.3353	0.2707	0.4264	0.4745	0.1984
Moderate	0.5803	0.5046	0.3408	0.2724	0.4294	0.4804	0.1969
Hard	0.5728	0.4926	0.3426	0.2724	0.4387	0.4811	0.2000
Average	0.5832	0.5104	0.3396	0.2718	0.4315	0.4787	0.1984

Color Quant

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.6087	0.5569	0.3317	0.2699	0.4283	0.4699	0.1984
Moderate	0.6040	0.5454	0.3336	0.2696	0.4225	0.4671	0.1940
Hard	0.5892	0.5192	0.3389	0.2715	0.4255	0.4754	0.1926
Average	0.6006	0.5405	0.3347	0.2703	0.4254	0.4708	0.1950

Motion Blur

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.6041	0.5446	0.3300	0.2698	0.4229	0.4667	0.1931
Moderate	0.5867	0.5114	0.3347	0.2711	0.4219	0.4693	0.1926
Hard	0.5796	0.5003	0.3362	0.2716	0.4281	0.4784	0.1911
Average	0.5901	0.5188	0.3336	0.2708	0.4243	0.4715	0.1923

Brightness

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.6112	0.5569	0.3310	0.2700	0.4204	0.4587	0.1923
Moderate	0.6073	0.5493	0.3310	0.2699	0.4244	0.4556	0.1926
Hard	0.6043	0.5427	0.3309	0.2697	0.4245	0.4538	0.1916
Average	0.6076	0.5497	0.3310	0.2699	0.4231	0.4560	0.1922

Low Light

Severity	NDS	mAP	mATE	mASE	mAOE	mAVE	mAAE
Easy	0.5904	0.5193	0.3344	0.2707	0.4147	0.4744	0.1983
Moderate	0.5792	0.4983	0.3375	0.2714	0.4145	0.4756	0.2001
Hard	0.5612	0.4671	0.3452	0.2731	0.4178	0.4854	0.2015
Average	0.5770	0.4949	0.3390	0.2717	0.4157	0.4785	0.2000

References

@article{chen2022autoalignv2,
  title={AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection},
  author={Chen, Zehui and Li, Zhenyu and Zhang, Shiquan and Fang, Liangji and Jiang, Qinhong and Zhao, Feng},
  journal={ECCV},
  year={2022}
}