Submission details of Team_MJM

Name Team_MJM
Paper Link N/A
Code Link N/A
AP 0.692043
Input N/A
Runtime 0.04 s
Environment 1 GPU (GTX 1080Ti)
Abstract we focus on exploring the robustness of the3D object detection in point clouds, which has been rarelydiscussed in existing approaches. We observe two crucialphenomena: 1) the detection accuracy of the hard objects,e.g., Pedestrians, is unsatisfactory, 2) when adding additionalnoise points, the performance of existing approaches de-creases rapidly. To alleviate these problems, a novel TANet isintroduced in this paper, which mainly contains a Triple At-tention (TA) module, and a Coarse-to-Fine Regression (CFR)module. By considering the channel-wise, point-wise andvoxel-wise attention jointly, the TA module enhances the cru-cial information of the target while suppresses the unsta-ble cloud points. Besides, the novel stacked TA further ex-ploits the multi-level feature attention. In addition, the CFRmodule boosts the accuracy of localization without excessivecomputation cost. Experimental results on the validation setof KITTI dataset demonstrate that, in the challenging noisycases, i.e., adding additional random noisy points around eachobject, the presented approach goes far beyond state-of-the-art approaches. Furthermore, for the 3D object detection taskof the KITTI benchmark, our approach ranks the first place onPedestrian class, by using the point clouds as the only input.The running speed is around 29 frames per second

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.785353
discovery-walk-2019-02-28_0 0.862199
discovery-walk-2019-02-28_1 0.707604
food-trucks-2019-02-12_0 0.821912
gates-ai-lab-2019-04-17_0 0.691086
gates-basement-elevators-2019-01-17_0 0.833876
gates-foyer-2019-01-17_0 0.811844
gates-to-clark-2019-02-28_0 0.722419
hewlett-class-2019-01-23_0 0.812821
hewlett-class-2019-01-23_1 0.914321
huang-2-2019-01-25_1 0.698431
huang-intersection-2019-01-22_0 0.71033
indoor-coupa-cafe-2019-02-06_0 0.598931
lomita-serra-intersection-2019-01-30_0 0.749483
meyer-green-2019-03-16_1 0.570703
nvidia-aud-2019-01-25_0 0.607331
nvidia-aud-2019-04-18_1 0.644652
nvidia-aud-2019-04-18_2 0.745971
outdoor-coupa-cafe-2019-02-06_0 0.605767
quarry-road-2019-02-28_0 0.697537
serra-street-2019-01-30_0 0.898434
stlc-111-2019-04-19_1 0.930737
stlc-111-2019-04-19_2 0.875322
tressider-2019-03-16_2 0.679754
tressider-2019-04-26_0 0.627377
tressider-2019-04-26_1 0.703741
tressider-2019-04-26_3 0.652244
total 0.692043