|
|
|
|
LEADER |
01524 am a22002293u 4500 |
001 |
137722 |
042 |
|
|
|a dc
|
100 |
1 |
0 |
|a Wang, Y
|e author
|
700 |
1 |
0 |
|a Fathi, A
|e author
|
700 |
1 |
0 |
|a Kundu, A
|e author
|
700 |
1 |
0 |
|a Ross, DA
|e author
|
700 |
1 |
0 |
|a Pantofaru, C
|e author
|
700 |
1 |
0 |
|a Funkhouser, T
|e author
|
700 |
1 |
0 |
|a Solomon, J
|e author
|
245 |
0 |
0 |
|a Pillar-Based Object Detection for Autonomous Driving
|
260 |
|
|
|b Springer International Publishing,
|c 2021-11-08T17:29:30Z.
|
856 |
|
|
|z Get fulltext
|u https://hdl.handle.net/1721.1/137722
|
520 |
|
|
|a © 2020, Springer Nature Switzerland AG. We present a simple and flexible object detection framework optimized for autonomous driving. Building on the observation that point clouds in this application are extremely sparse, we propose a practical pillar-based approach to fix the imbalance issue caused by anchors. In particular, our algorithm incorporates a cylindrical projection into multi-view feature learning, predicts bounding box parameters per pillar rather than per point or per anchor, and includes an aligned pillar-to-point projection module to improve the final prediction. Our anchor-free approach avoids hyperparameter search associated with past methods, simplifying 3D object detection while significantly improving upon state-of-the-art.
|
546 |
|
|
|a en
|
655 |
7 |
|
|a Article
|
773 |
|
|
|t 10.1007/978-3-030-58542-6_2
|
773 |
|
|
|t Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
|