Chensheng Peng

Chensheng Peng

About Me

I am a Ph.D. student at UC Berkeley, affiliated with Berkeley AI Research (BAIR) Lab and Berkeley DeepDrive (BDD).

My research focuses on 3D computer vision and its applications in Robotics.

Research Interests

Generation

3D/4D Generation
World Models

Reconstruction

Gaussian Splatting
NeRF-SLAM

Perception

Sensor Fusion
Multi-modal Learning

Publications

Selected research papers and preprints. * denotes equal contribution.

UNICST

UNICST: Next-scale Latent Prediction for Continuous Spatio-Temporal World Modeling

Yichen Xie*, Chensheng Peng*, Mazen Abdelfattah, Yihan Hu, Jiezhi Yang, Eric Higgins, Ryan Brigden, Wei Zhan
ICCV 2025 Workshop Oral @ Reliable and Interactable World Models (RIWM)

We introduce UNICST a unified 4D latent world model that jointly learns Continuous Spatio-Temporal representations with minimal inductive bias, enabling seamless, spatio-temporally coherent video generation.

Diff-GS

A Lesson in Splats: Teacher-Guided Diffusion for 3D Gaussian Splats Generation

Chensheng Peng, Ido Sobol, Masayoshi Tomizuka, Kurt Keutzer, Chenfeng Xu, Or Litany
ICCV 2025

We introduce a diffusion model for Gaussian Splats, SplatDiffusion, to enable generation of three-dimensional structures from single images. Our approach further incorporates a guidance mechanism to aggregate information from multiple views.

DeSiRe-GS

DeSiRe-GS: 4D Street Gaussians for Static-Dynamic Decomposition

Chensheng Peng*, Chengwei Zhang, Yixiao Wang, Chenfeng Xu, Yichen Xie, Wenzhao Zheng, Kurt Keutzer, Masayoshi Tomizuka
CVPR 2025

We present DeSiRe-GS, a self-supervised gaussian splatting representation, enabling effective static-dynamic decomposition and high-fidelity surface reconstruction in complex driving scenarios.

X-Drive

X-Drive: Cross-modality consistent multi-sensor data synthesis

Yichen Xie*, Chenfeng Xu*, Chensheng Peng, Shuqi Zhao, Nhat Ho, Alexander T. Pham, Mingyu Ding, Masayoshi Tomizuka, Wei Zhan
ICLR 2025

We propose a novel framework, X-DRIVE, to model the joint distribution of point clouds and multi-view images via a dual-branch latent diffusion model architecture.

Q-SLAM

Q-SLAM: Quadric Representations for Monocular SLAM

Chensheng Peng*, Chenfeng Xu*, Yue Wang, Mingyu Ding, Heng Yang, Masayoshi Tomizuka, Kurt Keutzer, Marco Pavone
CoRL 2024

In this study, we propose a novel approach that reimagines volumetric representations through the lens of quadric forms. We posit that most scene components can be effectively represented as quadric planes. Leveraging this assumption, we reshape the volumetric representations with million of cubes by several quadric planes, which leads to more accurate and efficient modeling of 3D scenes in SLAM contexts.

DELFlow

DELFlow: Dense Efficient Learning of Scene Flow for Large-Scale Point Clouds

Chensheng Peng, Guangming Wang, Xian Wan Lo, Xinrui Wu, Chenfeng Xu, Masayoshi Tomizuka, Wei Zhan, Hesheng Wang
ICCV 2023

Point clouds are naturally sparse, while image pixels are dense. We regularize raw points to a dense format by storing 3D coordinates in 2D grids.

InterMOT

Interactive multi-scale fusion of 2D and 3D features for multi-object vehicle tracking

Chensheng Peng*, Guangming Wang*, Yingying Gu, Jinpeng Zhang, Hesheng Wang
2023 IEEE T-ITS

In this paper, we propose multi-scale interactive query and fusion between pixel-wise and point-wise features to obtain more discriminative features. In addition, an attention mechanism is utilized to conduct soft feature fusion between multiple pixels and points to avoid inaccurate match problems.

PNAS-MOT

PNAS-MOT: Multi-Modal Object Tracking with Pareto Neural Architecture Search

Chensheng Peng, Zhaoyu Zeng, Jinling Gao, Jundong Zhou, Masayoshi Tomizuka, Xinbing Wang, Chenghu Zhou, Nanyang Ye
2024 IEEE RAL

We explore the use of the neural architecture search (NAS) methods to search for efficient architectures for tracking, aiming for low real-time latency while maintaining relatively high accuracy. We also propose a multi-modal framework to improve the robustness.

Get In Touch

Always open to discussing new ideas, collaborations, or research opportunities.

chensheng_peng [AT] berkeley [DOT] edu