Masked Autoencoder for Self-Supervised Pre-Training on Lidar Point Clouds
Paper i proceeding, 2023

Masked autoencoding has become a successful pretraining paradigm for Transformer models for text, images, and, recently, point clouds. Raw automotive datasets are suitable candidates for self-supervised pre-training as they generally are cheap to collect compared to annotations for tasks like 3D object detection (OD). However, the development of masked autoencoders for point clouds has focused solely on synthetic and indoor data. Consequently, existing methods have tailored their representations and models toward small and dense point clouds with homogeneous point densities. In this work, we study masked autoencoding for point clouds in an automotive setting, which are sparse and for which the point density can vary drastically among objects in the same scene. To this end, we propose Voxel-MAE, a simple masked autoencoding pre-training scheme designed for voxel representations. We pre-train the backbone of a Transformer-based 3D object detector to reconstruct masked voxels and to distinguish between empty and non-empty voxels. Our method improves the 3D OD performance by 1.75 mAP points and 1.05 NDS on the challenging nuScenes dataset. Further, we show that by pre-training with Voxel-MAE, we require only 40 of the annotated data to outperform a randomly initialized equivalent. Code is available at https://github.com/georghess/voxel-mae.

3d object detection

Self-supervised

Object detection

Voxel-MAE

Deep learning

Masked autoencoding

Författare

Georg Hess

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Zenseact AB

Johan Jaxing

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Elias Svensson

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

David Hagerman Olzon

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Christoffer Petersson

Zenseact AB

Chalmers, Matematiska vetenskaper, Algebra och geometri

Lennart Svensson

Chalmers, Elektroteknik, Signalbehandling och medicinsk teknik

Proceedings - 2023 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops, WACVW 2023

350-359
979-8-3503-2056-5 (ISBN)

IEEE Workshop on Applications of Computer Vision (WACV)
Waikoloa, USA,

Följning av objekt för självkörande fordon med hjälp av djup maskininlärning

Wallenberg AI, Autonomous Systems and Software Program, 2021-08-01 -- 2025-08-01.

Styrkeområden

Transport

Infrastruktur

C3SE (Chalmers Centre for Computational Science and Engineering)

Ämneskategorier

Datorseende och robotik (autonoma system)

DOI

10.1109/WACVW58289.2023.00039

Mer information

Senast uppdaterat

2023-07-19