Skip to content

Latest commit

 

History

History
155 lines (110 loc) · 10.1 KB

ORA3D.md

File metadata and controls

155 lines (110 loc) · 10.1 KB

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%. Operating points where recall or precision is less than $10$% are removed in order to minimize the impact of noise commonly seen in low precision and recall regions. If no operating point in this region is achieved, the AP for that class is set to zero. We then average over-matching thresholds of $\mathbb{D}={0.5, 1, 2, 4}$ 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 $10$%. If a $10$% recall is not achieved for a particular class, all TP errors for that class are set to $1$.

  • 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:

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

ORA3D-r101

Corruption NDS mAP mATE mASE mAOE mAVE mAAE
Clean 0.4436 0.3677 0.7319 0.2698 0.3890 0.8150 0.1975
Cam Crash 0.3055 0.1275 0.7952 0.2803 0.4549 0.8376 0.2145
Frame Lost 0.2750 0.0997 0.8362 0.3075 0.4963 0.8747 0.2340
Color Quant 0.3360 0.2382 0.8479 0.2848 0.5249 0.9516 0.2432
Motion Blur 0.2647 0.1527 0.8656 0.3497 0.6251 1.0433 0.3160
Brightness 0.4075 0.3252 0.7740 0.2741 0.4620 0.8372 0.2029
Low Light 0.2613 0.1509 0.8489 0.3445 0.6207 1.2113 0.3278
Fog 0.3959 0.3084 0.7822 0.2753 0.4515 0.8685 0.2048
Snow 0.1898 0.0757 0.9404 0.3857 0.7665 1.2890 0.3879

Experiment Log

Time:Time: Thu Jan 26 15:18:22 2023

Camera Crash

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3506 0.1950 0.7696 0.2739 0.4119 0.8088 0.2046
Moderate 0.2794 0.0882 0.8228 0.2826 0.4428 0.8784 0.2205
Hard 0.2865 0.0994 0.7933 0.2845 0.5101 0.8255 0.2183
Average 0.3055 0.1275 0.7952 0.2803 0.4549 0.8376 0.2145

Frame Lost

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3594 0.2174 0.7673 0.2743 0.4226 0.8302 0.1985
Moderate 0.2653 0.0696 0.8288 0.2904 0.4953 0.8713 0.2097
Hard 0.2003 0.0121 0.9124 0.3577 0.5711 0.9225 0.2937
Average 0.2750 0.0997 0.8362 0.3075 0.4963 0.8747 0.2340

Color Quant

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4194 0.3428 0.7709 0.2694 0.4063 0.8611 0.2117
Moderate 0.3496 0.2534 0.8246 0.2775 0.5054 0.9301 0.2331
Hard 0.2389 0.1185 0.9481 0.3074 0.6630 1.0636 0.2849
Average 0.3360 0.2382 0.8479 0.2848 0.5249 0.9516 0.2432

Motion Blur

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3844 0.2961 0.7830 0.2799 0.4802 0.8804 0.2134
Moderate 0.2348 0.1021 0.9026 0.3154 0.6737 1.0273 0.2714
Hard 0.1750 0.0599 0.9111 0.4538 0.7214 1.2221 0.4632
Average 0.2647 0.1527 0.8656 0.3497 0.6251 1.0433 0.3160

Brightness

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4398 0.3612 0.7458 0.2691 0.3976 0.7939 0.2019
Moderate 0.4072 0.3261 0.7735 0.2728 0.4613 0.8455 0.2050
Hard 0.3756 0.2882 0.8028 0.2805 0.5272 0.8721 0.2019
Average 0.4075 0.3252 0.7740 0.2741 0.4620 0.8372 0.2029

Low Light

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3278 0.2288 0.8060 0.2837 0.5339 1.0152 0.2426
Moderate 0.2731 0.1560 0.8631 0.3024 0.6129 1.1647 0.2702
Hard 0.1829 0.0680 0.8776 0.4473 0.7152 1.4539 0.4706
Average 0.2613 0.1509 0.8489 0.3445 0.6207 1.2113 0.3278

Fog

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4157 0.3308 0.7635 0.2717 0.4238 0.8362 0.2022
Moderate 0.3982 0.3125 0.7831 0.2750 0.4508 0.8667 0.2048
Hard 0.3740 0.2819 0.7999 0.2793 0.4800 0.9027 0.2075
Average 0.3959 0.3084 0.7822 0.2753 0.4515 0.8685 0.2048

Snow

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.2457 0.1197 0.9051 0.3191 0.6306 1.2343 0.2871
Moderate 0.1668 0.0584 0.9440 0.4163 0.8276 1.2921 0.4359
Hard 0.1570 0.0492 0.9721 0.4217 0.8413 1.3406 0.4408
Average 0.1898 0.0757 0.9404 0.3857 0.7665 1.2890 0.3879

References

@article{roh2022ora3d,
  title={Ora3d: Overlap region aware multi-view 3d object detection},
  author={Roh, Wonseok and Chang, Gyusam and Moon, Seokha and Nam, Giljoo and Kim, Chanyoung and Kim, Younghyun and Kim, Sangpil and Kim, Jinkyu},
  journal={arXiv preprint arXiv:2207.00865},
  year={2022}
}