Detection

The MPD dataset mainly covers complex road scenes, such as road shadows, camera overexposure, rain, fog, lane contamination, etc. The robustness of the autonomous driving algorithm is precisely tested by these scenes. Algorithms that perform well on other related datasets may not perform well on MPD, whereas algorithms that perform well on MPD will be more likely to have strong adaptability in real scenes. Researchers in the field of autonomous driving are welcome to evaluate the performance of algorithms based on MPD!.

The OpenMPD dataset mainly covers complex road scenes, such as road shadows, camera overexposure, rain, fog, lane contamination, etc. The robustness of the autonomous driving algorithm is precisely tested by these scenes. All those algorithms that have great performance on other datasets may not work well on OpenMPD. We encourage researchers to try to test and train their algorithms on OpenMPD. So that, those algorithms can face the real challenge on the true environment. If you have a good performance on OpenMPD, you are welcome to submit your code and algorithm to our website, we will put you on the leaderboard to let you know how good you are comparing to others.

Metrics for 2D object detection

mean Average Precision. mAP is a popular metric in measuring the accuracy of object detectors's performance. mAP is calculated by Precision and Recall , and both of them can be calculated by True Positive(TP) 、 False Positive(FP) 、False Negative (FN)、True Negative (TN) etc.

图片1.png

An intersection Over Union (IOU) threshold Predefined determines whether the prediction is a True Positive(TP) or a False Positive(FP). PASCAL VOC is a dataset in object detection competition. For the PASCAL VOC challenge, a prediction is positive if IoU  0.5.

The general definition for the Average Precision (AP) is finding the area under the precision-recall curve above.

图片2.png

We will use above metrics to evaluate your algorithm. The full information for this content will be on our paper.

Cars
2D Object Detection

  • Method

  • Code

  • Moderate

  • Runtime

  • Environment

  • Article

  • 1

  • PC-CNN-V2


  • 95.20%

  • 0.5s

  • GPU@2.5Ghz


  • 2

  • F-PointNet


  • 95.17%

  • 0.17s

  • GPU@3.0Ghz


3D Object Detection

  • Method

  • Code

  • Moderate

  • Runtime

  • Environment

  • Article

  • 1

  • PC-CNN-V2


  • 95.20%

  • 0.5s

  • GPU@2.5Ghz


  • 2

  • F-PointNet


  • 95.17%

  • 0.17s

  • GPU@3.0Ghz


Pedestrian
2D Object Detection

  • Method

  • Code

  • Moderate

  • Runtime

  • Environment

  • Article

  • 1

  • PC-CNN-V2


  • 95.20%

  • 0.5s

  • GPU@2.5Ghz


  • 2

  • F-PointNet


  • 95.17%

  • 0.17s

  • GPU@3.0Ghz


3D Object Detection

  • Method

  • Code

  • Moderate

  • Runtime

  • Environment

  • Article

  • 1

  • PC-CNN-V2


  • 95.20%

  • 0.5s

  • GPU@2.5Ghz


  • 2

  • F-PointNet


  • 95.17%

  • 0.17s

  • GPU@3.0Ghz


Tsinghua University

Haidian District, Beijing, 100084, P. R. China

Email:xyzhang@tsinghua.edu.cn

© Copyright 2016-2020 www.OpenMPD.com. All Rights Reserved