Skip to content

Latest commit

 

History

History
153 lines (110 loc) · 9.14 KB

PETR-vov.md

File metadata and controls

153 lines (110 loc) · 9.14 KB
Raw

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:

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

PETR-VovNet

Corruption NDS mAP mATE mASE mAOE mAVE mAAE
Clean 0.4550 0.4035 0.7362 0.2710 0.4316 0.8249 0.2039
Cam Crash 0.2924 0.1408 0.8167 0.2854 0.5492 0.9014 0.2267
Frame Lost 0.2792 0.1153 0.8311 0.2909 0.5662 0.8816 0.2144
Color Quant 0.2968 0.2089 0.8818 0.3455 0.5997 1.0875 0.3123
Motion Blur 0.2490 0.1395 0.9521 0.3153 0.7424 1.0353 0.2639
Brightness 0.3858 0.3199 0.7982 0.2779 0.5256 0.9342 0.2112
Low Light 0.2305 0.1221 0.8897 0.3645 0.6960 1.2311 0.3553
Fog 0.3703 0.2815 0.8337 0.2778 0.4982 0.8833 0.2111
Snow 0.2632 0.1653 0.8980 0.3138 0.7034 1.1314 0.2886

Experiment Log

Time: Fri Jan 20 23:39:21 2023

Camera Crash

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3348 0.2107 0.8052 0.2819 0.5253 0.8717 0.2216
Moderate 0.2630 0.0998 0.8451 0.2836 0.5406 0.9566 0.2431
Hard 0.2795 0.1118 0.7998 0.2907 0.5817 0.8759 0.2155
Average 0.2924 0.1408 0.8167 0.2854 0.5492 0.9014 0.2267

Frame Lost

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3619 0.2459 0.7765 0.2761 0.4816 0.8678 0.2083
Moderate 0.2618 0.0828 0.8323 0.2908 0.5614 0.8974 0.2143
Hard 0.2140 0.0171 0.8846 0.3059 0.6556 0.8795 0.2205
Average 0.2792 0.1153 0.8311 0.2909 0.5662 0.8816 0.2144

Color Quant

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4254 0.3647 0.7639 0.2748 0.4584 0.8558 0.2171
Moderate 0.3156 0.2212 0.8591 0.2954 0.5736 0.9774 0.2447
Hard 0.1495 0.0408 1.0224 0.4662 0.7670 1.4292 0.4752
Average 0.2968 0.2089 0.8818 0.3455 0.5997 1.0875 0.3123

Motion Blur

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3826 0.3023 0.8319 0.2758 0.5137 0.8587 0.2049
Moderate 0.2067 0.0793 0.9682 0.3181 0.7665 1.0943 0.2767
Hard 0.1575 0.0369 1.0563 0.3519 0.9471 1.1529 0.3102
Average 0.2490 0.1395 0.9521 0.3153 0.7424 1.0353 0.2639

Brightness

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4348 0.3750 0.7586 0.2718 0.4595 0.8336 0.2035
Moderate 0.3785 0.3121 0.8002 0.2788 0.5339 0.9532 0.2099
Hard 0.3441 0.2726 0.8357 0.2830 0.5835 1.0159 0.2202
Average 0.3858 0.3199 0.7982 0.2779 0.5256 0.9342 0.2112

Low Light

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.2988 0.1925 0.8503 0.2944 0.5740 1.1010 0.2562
Moderate 0.2427 0.1204 0.8879 0.3274 0.6773 1.2340 0.2825
Hard 0.1500 0.0533 0.9308 0.4716 0.8367 1.3582 0.5271
Average 0.2305 0.1221 0.8897 0.3645 0.6960 1.2311 0.3553

Fog

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3956 0.3151 0.8094 0.2757 0.4726 0.8566 0.2052
Moderate 0.3686 0.2812 0.8331 0.2785 0.5047 0.8900 0.2131
Hard 0.3468 0.2482 0.8585 0.2793 0.5173 0.9033 0.2149
Average 0.3703 0.2815 0.8337 0.2778 0.4982 0.8833 0.2111

Snow

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3494 0.2715 0.8262 0.2833 0.5439 0.9719 0.2380
Moderate 0.2312 0.1232 0.9234 0.3179 0.7635 1.1908 0.2995
Hard 0.2090 0.1012 0.9443 0.3401 0.8029 1.2316 0.3283
Average 0.2632 0.1653 0.8980 0.3138 0.7034 1.1314 0.2886

References

@article{liu2022petr,
  title = {PETR: Position Embedding Transformation for Multi-View 3D Object Detection},
  author = {Liu, Yingfei and Wang, Tiancai and Zhang, Xiangyu and Sun, Jian},
  journal = {arXiv preprint arXiv:2203.05625},
  year = {2022},
}