Skip to content

Latest commit

 

History

History
154 lines (110 loc) · 9.26 KB

SOLOFusion-Long.md

File metadata and controls

154 lines (110 loc) · 9.26 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:

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

SOLOFusion-Long-Only

Corruption NDS mAP mATE mASE mAOE mAVE mAAE
Clean 0.4850 0.3862 0.6292 0.2840 0.6387 0.3151 0.2141
Cam Crash 0.3159 0.1173 0.7462 0.2938 0.6939 0.4614 0.2327
Frame Lost 0.2490 0.1121 0.7824 0.3529 0.8133 0.8167 0.3249
Color Quant 0.3598 0.2233 0.7704 0.3206 0.7326 0.4266 0.2681
Motion Blur 0.3460 0.1969 0.7765 0.2973 0.7849 0.4262 0.2395
Brightness 0.4002 0.2726 0.7163 0.3113 0.6754 0.4251 0.2328
Low Light 0.2814 0.1301 0.7669 0.3701 0.7913 0.5548 0.3534
Fog 0.3991 0.2570 0.7230 0.2947 0.7084 0.3678 0.2002
Snow 0.1480 0.0590 0.8901 0.5666 0.9179 0.7932 0.6480

Experiment Log

Time: Thu Apr 6 15:37:12 2023

Camera Crash

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3654 0.1912 0.7037 0.2866 0.6913 0.4032 0.2176
Moderate 0.2961 0.0833 0.8047 0.2879 0.6726 0.4534 0.2368
Hard 0.2861 0.0775 0.7301 0.3070 0.7178 0.5276 0.2438
Average 0.3159 0.1173 0.7462 0.2938 0.6939 0.4614 0.2327

Frame Lost

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3804 0.2524 0.6895 0.2951 0.7106 0.5281 0.2355
Moderate 0.2239 0.0723 0.8009 0.3184 0.8774 0.8632 0.2625
Hard 0.1427 0.0115 0.8568 0.4451 0.8519 1.0589 0.4766
Average 0.2490 0.1121 0.7824 0.3529 0.8133 0.8167 0.3249

Color Quant

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4541 0.3436 0.6797 0.2846 0.6621 0.3287 0.2217
Moderate 0.3806 0.2343 0.7579 0.2941 0.7034 0.3848 0.2249
Hard 0.2446 0.0918 0.8736 0.3831 0.8324 0.5662 0.3578
Average 0.3598 0.2233 0.7704 0.3206 0.7326 0.4266 0.2681

Motion Blur

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4371 0.3207 0.6822 0.2876 0.7112 0.3401 0.2118
Moderate 0.3227 0.1592 0.7891 0.2963 0.7985 0.4414 0.2432
Hard 0.2783 0.1109 0.8581 0.3081 0.8450 0.4970 0.2635
Average 0.3460 0.1969 0.7765 0.2973 0.7849 0.4262 0.2395

Brightness

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4501 0.3321 0.6740 0.2890 0.6555 0.3415 0.1992
Moderate 0.3998 0.2642 0.7303 0.2993 0.6977 0.3963 0.1992
Hard 0.3507 0.2215 0.7446 0.3456 0.6731 0.5376 0.3001
Average 0.4002 0.2726 0.7163 0.3113 0.6754 0.4251 0.2328

Low Light

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3408 0.1831 0.7481 0.3042 0.7764 0.4396 0.2389
Moderate 0.2859 0.1401 0.7622 0.3786 0.8201 0.5294 0.3511
Hard 0.2176 0.0673 0.7903 0.4276 0.7773 0.6955 0.4701
Average 0.2814 0.1301 0.7669 0.3701 0.7913 0.5548 0.3534

Fog

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4132 0.2766 0.7144 0.2942 0.6912 0.3548 0.1966
Moderate 0.3985 0.2580 0.7213 0.2941 0.7152 0.3705 0.2042
Hard 0.3856 0.2364 0.7334 0.2959 0.7188 0.3780 0.1997
Average 0.3991 0.2570 0.7230 0.2947 0.7084 0.3678 0.2002

Snow

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.2672 0.1224 0.7825 0.3796 0.8374 0.5530 0.3871
Moderate 0.0931 0.0295 0.9408 0.6561 0.9475 0.8910 0.7811
Hard 0.0835 0.0253 0.9469 0.6640 0.9687 0.9357 0.7758
Average 0.1480 0.0590 0.8901 0.5666 0.9179 0.7932 0.6480

References

@article{Park2022TimeWT,
  title={Time Will Tell: New Outlooks and A Baseline for Temporal Multi-View 3D Object Detection},
  author={Park, Jinhyung and Xu, Chenfeng and Yang, Shijia and Keutzer, Kurt and Kitani, Kris and Tomizuka, Masayoshi and Zhan, Wei},
  booktitle={International Conference on Learning Representations},
  year={2023}
}