Submission details of EPNet++

Name EPNet++
Paper Link https://arxiv.org/pdf/2007.08856.pdf
Code Link https://github.com/happinesslz/EPNet
AP 0.666313
Input N/A
Runtime 0.2 s
Environment 1 GPU(Titan V)
Abstract In this paper, we aim at addressing two critical issues in the 3D detection task, including the exploitation of multiple sensors (namely LiDAR point cloud and camera image), as well as the inconsistency be- tween the localization and classification confidence. To this end, we pro- pose a novel fusion module to enhance the point features with semantic image features in a point-wise manner. Besides, a consistency forcing loss is employed to explicitly encourage the consistency of both the localiza- tion and classification confidence. We design an end-to-end learnable framework named EPNet to integrate these two components. Extensive experiments on the KITTI and SUN-RGBD datasets demonstrate the superiority of EPNet over the state-of-the-art methods.

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.806739
discovery-walk-2019-02-28_0 0.922893
discovery-walk-2019-02-28_1 0.840583
food-trucks-2019-02-12_0 0.849237
gates-ai-lab-2019-04-17_0 0.760519
gates-basement-elevators-2019-01-17_0 0.834842
gates-foyer-2019-01-17_0 0.889946
gates-to-clark-2019-02-28_0 0.925571
hewlett-class-2019-01-23_0 0.792839
hewlett-class-2019-01-23_1 0.95852
huang-2-2019-01-25_1 0.71466
huang-intersection-2019-01-22_0 0.823571
indoor-coupa-cafe-2019-02-06_0 0.611619
lomita-serra-intersection-2019-01-30_0 0.905275
meyer-green-2019-03-16_1 0.626977
nvidia-aud-2019-01-25_0 0.652813
nvidia-aud-2019-04-18_1 0.680664
nvidia-aud-2019-04-18_2 0.854472
outdoor-coupa-cafe-2019-02-06_0 0.650742
quarry-road-2019-02-28_0 0.815685
serra-street-2019-01-30_0 0.891822
stlc-111-2019-04-19_1 0.928024
stlc-111-2019-04-19_2 0.862916
tressider-2019-03-16_2 0.861519
tressider-2019-04-26_0 0.560883
tressider-2019-04-26_1 0.568443
tressider-2019-04-26_3 0.59913
total 0.666313