Skip to content

Latest commit

 

History

History
54 lines (38 loc) · 4.65 KB

README.md

File metadata and controls

54 lines (38 loc) · 4.65 KB
Raw

Second: Sparsely embedded convolutional detection

SECOND: Sparsely Embedded Convolutional Detection

Abstract

LiDAR-based or RGB-D-based object detection is used in numerous applications, ranging from autonomous driving to robot vision. Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point cloud LiDAR data. However, problems remain, including a slow inference speed and low orientation estimation performance. We therefore investigate an improved sparse convolution method for such networks, which significantly increases the speed of both training and inference. We also introduce a new form of angle loss regression to improve the orientation estimation performance and a new data augmentation approach that can enhance the convergence speed and performance. The proposed network produces state-of-the-art results on the KITTI 3D object detection benchmarks while maintaining a fast inference speed.

Introduction

We implement SECOND and provide the results and checkpoints on KITTI dataset.

Results and models

KITTI

Backbone Class Lr schd Mem (GB) Inf time (fps) mAP Download
SECFPN Car cyclic 80e 5.4 79.07 model | log
SECFPN (FP16) Car cyclic 80e 2.9 78.72 model| log
SECFPN 3 Class cyclic 80e 5.4 65.74 model | log
SECFPN (FP16) 3 Class cyclic 80e 2.9 67.4 model | log

Waymo

Backbone Load Interval Class Lr schd Mem (GB) Inf time (fps) mAP@L1 mAPH@L1 mAP@L2 mAPH@L2 Download
SECFPN 5 3 Class 2x 8.12 65.3 61.7 58.9 55.7 log
above @ Car 2x 8.12 67.1 66.6 58.7 58.2
above @ Pedestrian 2x 8.12 68.1 59.1 59.5 51.5
above @ Cyclist 2x 8.12 60.7 59.5 58.4 57.3

Note:

  • See more details about metrics and data split on Waymo HERE. For implementation details, we basically follow the original settings. All of these results are achieved without bells-and-whistles, e.g. ensemble, multi-scale training and test augmentation.
  • FP16 means Mixed Precision (FP16) is adopted in training.

Citation

@article{yan2018second,
  title={Second: Sparsely embedded convolutional detection},
  author={Yan, Yan and Mao, Yuxing and Li, Bo},
  journal={Sensors},
  year={2018},
  publisher={Multidisciplinary Digital Publishing Institute}
}