arrow
Return

Enhanced Spatiotemporal Consistency for Image-to-LiDAR Data Pretraining

delete2025-12-05
delete0
PRE
AI
X
Xiang Xu
L
Lingdong Kong
H
Hui Shuai
W
Wenwei Zhang
L
Liang Pan
陈
陈朝宇 (Kai Chen)
Z
Ziwei Liu
刘青山 cover
刘青山 (Qingshan Liu)
DOI:10.1109/TPAMI.2025.3640589delete
deleteOriginal
deleteShare
deleteSave
Abstract

Abstract

En 中文
LiDAR representation learning has emerged as a promising approach to reducing reliance on costly and labor-intensive human annotations. While existing methods primarily focus on spatial alignment between LiDAR and camera sensors, they often overlook the temporal dynamics critical for capturing motion and scene continuity in driving scenarios. To address this limitation, we propose SuperFlow++, a novel framework that integrates spatiotemporal cues in both pretraining and downstream tasks using consecutive LiDAR-camera pairs. SuperFlow++ introduces four key components: (1) a view consistency alignment module to unify semantic information across camera views, (2) a dense-to-sparse consistency regularization mechanism to enhance feature robustness across varying point cloud densities, (3) a flow-based contrastive learning approach that models temporal relationships for improved scene understanding, and (4) a temporal voting strategy that propagates semantic information across LiDAR scans to improve prediction consistency. Extensive evaluations on 11 heterogeneous LiDAR datasets demonstrate that SuperFlow++ outperforms state-of-the-art methods across diverse tasks and driving conditions. Furthermore, by scaling both 2D and 3D backbones during pretraining, we uncover emergent properties that provide deeper insights into developing scalable 3D foundation models. With strong generalizability and computational efficiency, SuperFlow++ establishes a new benchmark for data-efficient LiDAR-based perception in autonomous driving.
Keywords:
Autonomous driving
LiDAR segmentation
3D object detection
data pretraining
cross-sensor contrastive learning

Journal

IEEE Transactions on Pattern Analysis and Machine Intelligence cover
IEEE Transactions on Pattern Analysis and Machine Intelligence
IF:
18.6
Papers:
2.1K
Citations:
9.8W

Organization

N
national university of singapore and cnrs
Scholars:
1
Papers: 1
Citations: 0
N
nanjing university of posts and telecommunications
Scholars:
3.8K
Papers: 1.6K
Citations: 0
N
nanyang technological university
Scholars:
2.5K
Papers: 1.6K
Citations: 1
S
Shanghai AI Laboratory
Scholars:
18
Papers: 11
Citations: 0
N
Nanjing University of Aeronautics and Astronautics
Scholars:
7.4K
Papers: 3.1K
Citations: 2.4W
researcher View more organizations
Cited Papers

Cited Papers

FRNet: Frustum-Range Networks for Scalable LiDAR Segmentation
err2025-01-01
err0
PREAI
errXu,Xiang; Kong,Lingdong; Shuai,Hui; Liu,Qingshan
errShare
errSave
Contrastive Representation Learning: A Framework and Review
err2020-01-01
err517
errOAAI
errLe-Khac, Phuc H.; Healy, Graham; Smeaton, Alan F.
errShare
errSave
err
IF0
err
err0
PREAI
err
errShare
errSave
Image-to-Lidar Self-Supervised Distillation for Autonomous Driving Data
err2022-06-01
err0
errOAAI
errCorentin Sautier; Gilles Puy; Spyros Gidaris; Alexandre Boulch; Andrei Bursuc; Renaud Marlet
errShare
errSave
errShare
errSave
MAELi: Masked Autoencoder for Large-Scale LiDAR Point Clouds
err
IF0
err2024-01-03
err0
PREAI
errGeorg Krispel; David Schinagl; Christian Fruhwirth-Reisinger; Horst Possegger; Horst Bischof
errShare
errSave
Three Pillars Improving Vision Foundation Model Distillation for Lidar
err2024-06-16
err0
PREAI
errPuy,Gilles; Gidaris,Spyros; Boulch,Alexandre; Siméoni,Oriane; Sautier,Corentin; Pérez,Patrick; Bursuc,Andrei; Marlet,Renaud
errShare
errSave
BEVContrast: Self-Supervision in BEV Space for Automotive Lidar Point Clouds
err2024-03-18
err0
errOAAI
errCorentin Sautier; Gilles Puy; Alexandre Boulch; Renaud Marlet; Vincent Lepetit
errShare
errSave
Exploring Geometry-aware Contrast and Clustering Harmonization for Self-supervised 3D Object Detection
err2021-10-01
err0
PREAI
errHanxue Liang; Chenhan Jiang; Dapeng Feng; Xin Chen; Hang Xu; Xiaodan Liang; Wei Zhang; Zhenguo Li; Luc Van Gool
errShare
errSave
researcher View more