Chenjie Cao ζΉθΎ°ζ·
Researcher at Tencent Hunyuan3D. Ph.D. from Fudan University supervised by
Prof. Yanwei Fu.
Research interests span image inpainting & synthesis,
multi-view stereo, 3D reconstruction & generation,
and video generation.
Visitors:
News π
2026.07
HY-World 2.1 is now available
Team HY-World's July 2026 update brings sharper, more detailed panorama generation; fully in-house reconstruction with more accurate panoramic depth; a gallery mode; normal-regularized 3DGS optimization; and broader view completion at no additional runtime.
2026.07
OmniX was accepted to ECCV 2026.
2026.04
Two papers were accepted to ICML 2026: WorldCompass and RealisMotion.
2026.04
HY-World 2.0 was released as a multi-modal 3D world.
2026.02
WorldStereo was accepted to CVPR 2026.
2025.11
EarthCrafter was accepted to AAAI 2026 as an Oral presentation.
2025.8
Uni3C was accepted to SIGGRAPH Asia 2025.
Experience
2025.7 β present
Researcher, Tencent Hunyuan3D
2024.1 β 2025.7
Researcher, DAMO Academy, Alibaba Group
2024 β present
Post-doc, Computer Science and Technology, Fudan University
2020 β 2024
Ph.D. in Statistics (Machine Learning), Fudan University
2019 β 2020
Algorithm Engineer, PingAn OneConnect
2016 β 2019
M.Sc. in Computer Science, ECUST
2012 β 2016
B.Sc. in Computer Science, ECUST
Publications
* denotes equal contribution
2026
ECCV 2026
A feed-forward 4D reconstruction framework that predicts dense 3D point trajectories from videos with large camera motion, enabled by dynamic tokens and a large-scale 4D data engine.
Arxiv 2026
A multi-modal world model framework that unifies world generation and world reconstruction, producing navigable 3D scenes from text, images, or videos.
ICML 2026
A novel RL post-training framework for long-horizon, interactive video-based world models.
ICML 2026
A world-space video generation framework with decomposed human motion control and a Trajectory100 evaluation set.
CVPR 2026
A novel framework that bridges camera-guided video generation and 3D reconstruction via global-geometric memory and spatial-stereo memory.
AAAI 2026 Oral
A geographic 3D generation framework for extremely large-scale 3D generation, with the largest 3D aerial dataset (50k scenes, 45M frames).
2025
SIGGRAPH Asia 2025
A unified 3D-enhanced framework for precise control of both camera and human motion in video generation via PCDController and SMPL-X integration.
ICCV 2025
Rethinking LoRA fusion through linear scalability, orthogonality, and norm consistency for video control.
CVPR 2025
A multi-view diffusion model enhanced with 3D priors warped using metric depth and camera poses for versatile Novel View Synthesis.
CVPR 2025 Highlight
Leveraging MAE priors to stabilize StableDiffusion/FLUX inpainting with an effective augmentation strategy to eliminate chromatic aberration.
2024
NeurIPS 2024
Re-formulating 3D editing as a multi-view 2D inpainting task with reference guidance.
NeurIPS 2024
A framework for animating any static 3D model with a large-scale multi-view video dataset (MV-Video).
ECCV 2024
A Variance alignment based 3D-2D Collaborative Denoising framework for texture synthesis.
ECCV 2024
An efficient sparse-controlled video-to-4D framework that decouples motion and appearance.
ECCV 2024
Comprehensively exploring multi-view feature priors from seven pretext visual tasks to strengthen Neural Surface Reconstruction.
CVPR 2024
Formulating reference-guided multi-view synthesis as contextual inpainting via prompt tuning on pre-trained T2I models.
TMLR 2024
A single diffusion generative model addressing various sub-tasks with task-specific prompts.
ICLR 2024
Prudently maximizing the inherent characteristics of attention to enhance various components of the MVS pipeline.
2023
ICCV 2023
A transformer-based cascaded matching model with NMS filter for spatially informative keypoint matching.
ACM MM 2023
A Siamese network with a reciprocal loss for inlier/outlier classification in two-view matching.
TPAMI 2023
Extension of ZITS with learning-based edges (L-Edge) as new priors to enhance meaningful structure recovery.
CVPR 2023
Improving optical flow estimation through image matching pre-text task.
TMLR 2023
Pre-trained ViTs for MVS with unified classification and regression depth predictions. 1st on Tanks-and-Temples, 2nd on GigaMVS 2022.
2022
ECCV 2022
Image inpainting with Masked AutoEncoder pre-training and attention-based restoration.
CVPR 2022
Learning structural priors (lines, edges) with transformers, then recovering textures with FFC-based CNNs.
TPAMI 2022
Extension of P2M++ that also works on SDF-based methods.
ICASSP 2022
Leveraging GAN inversion and differentiable edge detection for effective face editing.
2021
NeurIPS 2021
Local autoregressive model for local editing with two-stream convolution to tackle information leakage.
ICCV 2021
First work to introduce segment lines to improve image inpainting.
2020 & Earlier
COLING 2020
A comprehensive Chinese NLP benchmark.
TNNLS 2020
Ensemble learning approach to address imbalanced problems.
LREC 2020
Improving BERT pre-training with Sentence Insertion. 1st place of CMRC2019.
Neural Networks 2019
Ensemble learning for imbalanced problems.
KBS 2018
Cost-sensitive stacking learning with feature inverse mapping for imbalanced problems.
ASC 2018
Using SVDD to address imbalanced problems.
Preprint
First introduction of wavelet prior for image inpainting with axial-transformer for face/structural recovery.