Submission details of TANet

Name TANet
Paper Link https://arxiv.org/abs/1912.05163
Code Link https://github.com/happinesslz/TANet
AP 0.549364
Input 3d
Runtime 0.28 s
Environment 1 GPU (Titan Tesla K40c)
Abstract TANet is based on Pointpillars, and it includes three types of attention(voxelwise, pointwise and channelwise).

Detailed results

Overall precision/recall curve

Per-sequence results

sequence name AP
cubberly-auditorium-2019-04-22_1 0.704401
discovery-walk-2019-02-28_0 0.753611
discovery-walk-2019-02-28_1 0.769257
food-trucks-2019-02-12_0 0.638266
gates-ai-lab-2019-04-17_0 0.596082
gates-basement-elevators-2019-01-17_0 0.743445
gates-foyer-2019-01-17_0 0.715264
gates-to-clark-2019-02-28_0 0.791268
hewlett-class-2019-01-23_0 0.721525
hewlett-class-2019-01-23_1 0.782696
huang-2-2019-01-25_1 0.555615
huang-intersection-2019-01-22_0 0.61657
indoor-coupa-cafe-2019-02-06_0 0.39306
lomita-serra-intersection-2019-01-30_0 0.806402
meyer-green-2019-03-16_1 0.462287
nvidia-aud-2019-01-25_0 0.420534
nvidia-aud-2019-04-18_1 0.507312
nvidia-aud-2019-04-18_2 0.550092
outdoor-coupa-cafe-2019-02-06_0 0.408209
quarry-road-2019-02-28_0 0.766789
serra-street-2019-01-30_0 0.729263
stlc-111-2019-04-19_1 0.829929
stlc-111-2019-04-19_2 0.703237
tressider-2019-03-16_2 0.779034
tressider-2019-04-26_0 0.481487
tressider-2019-04-26_1 0.566074
tressider-2019-04-26_3 0.504245
total 0.549364