Skip to content

Latest commit

 

History

History
154 lines (110 loc) · 9.26 KB

SOLOFusion.md

File metadata and controls

154 lines (110 loc) · 9.26 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:

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

SOLOFusion CBGS

Corruption NDS mAP mATE mASE mAOE mAVE mAAE
Clean 0.5381 0.4299 0.5842 0.2747 0.4564 0.2426 0.2103
Cam Crash 0.3806 0.1590 0.6607 0.2773 0.5186 0.3152 0.2176
Frame Lost 0.3464 0.1671 0.7161 0.3042 0.5557 0.5292 0.2668
Color Quant 0.4058 0.2572 0.6910 0.3200 0.6217 0.3434 0.2514
Motion Blur 0.3642 0.2019 0.7191 0.3244 0.6643 0.3834 0.2762
Brightness 0.4329 0.2959 0.6532 0.3238 0.5353 0.3808 0.2577
Low Light 0.2626 0.1237 0.7258 0.4567 0.7598 0.5910 0.4597
Fog 0.4480 0.2923 0.6502 0.2883 0.5496 0.2958 0.1973
Snow 0.1376 0.0561 0.8722 0.6480 0.8219 0.8363 0.7255

Experiment Log

Time: Thu Apr 6 15:32:35 2023

Camera Crash

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4354 0.2507 0.6388 0.2746 0.4974 0.2833 0.2050
Moderate 0.3589 0.1192 0.7112 0.2740 0.4860 0.3153 0.2203
Hard 0.3474 0.1073 0.6322 0.2833 0.5724 0.3470 0.2275
Average 0.3806 0.1590 0.6607 0.2773 0.5186 0.3152 0.2176

Frame Lost

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4695 0.3199 0.6321 0.2771 0.4718 0.3071 0.2166
Moderate 0.3349 0.1381 0.7287 0.2907 0.5627 0.5172 0.2427
Hard 0.2348 0.0434 0.7874 0.3449 0.6325 0.7632 0.3410
Average 0.3464 0.1671 0.7161 0.3042 0.5557 0.5292 0.2668

Color Quant

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.5153 0.3935 0.6014 0.2793 0.4698 0.2545 0.2095
Moderate 0.4339 0.2723 0.6643 0.2910 0.5546 0.2960 0.2170
Hard 0.2683 0.1057 0.8073 0.3897 0.8408 0.4796 0.3278
Average 0.4058 0.2572 0.6910 0.3200 0.6217 0.3434 0.2514

Motion Blur

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4856 0.3532 0.6204 0.2860 0.5388 0.2626 0.2020
Moderate 0.3410 0.1527 0.7529 0.3018 0.6926 0.3655 0.2411
Hard 0.2661 0.0999 0.7841 0.3854 0.7615 0.5222 0.3855
Average 0.3642 0.2019 0.7191 0.3244 0.6643 0.3834 0.2762

Brightness

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.5035 0.3723 0.6119 0.2795 0.4692 0.2770 0.1891
Moderate 0.4451 0.2871 0.6555 0.2816 0.5303 0.3281 0.1890
Hard 0.3500 0.2282 0.6923 0.4103 0.6063 0.5374 0.3951
Average 0.4329 0.2959 0.6532 0.3238 0.5353 0.3808 0.2577

Low Light

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.3339 0.1807 0.6933 0.3743 0.7006 0.4598 0.3370
Moderate 0.2732 0.1287 0.6953 0.4181 0.7404 0.6177 0.4400
Hard 0.1806 0.0618 0.7889 0.5777 0.8383 0.6955 0.6022
Average 0.2626 0.1237 0.7258 0.4567 0.7598 0.5910 0.4597

Fog

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.4656 0.3199 0.6426 0.2825 0.5436 0.2817 0.1932
Moderate 0.4505 0.2952 0.6465 0.2878 0.5426 0.2970 0.1972
Hard 0.4280 0.2618 0.6614 0.2946 0.5626 0.3088 0.2014
Average 0.4480 0.2923 0.6502 0.2883 0.5496 0.2958 0.1973

Snow

Severity NDS mAP mATE mASE mAOE mAVE mAAE
Easy 0.2638 0.1110 0.7757 0.4273 0.6741 0.5913 0.4489
Moderate 0.0685 0.0301 0.9115 0.7930 0.8853 0.9709 0.9051
Hard 0.0807 0.0272 0.9295 0.7238 0.9064 0.9468 0.8226
Average 0.1376 0.0561 0.8722 0.6480 0.8219 0.8363 0.7255

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}
}