Weichen Zhang

About Me

I am a highly motivated Machine Learning and Deep Learning Researcher and Engineer, currently a Postdoctoral Researcher at the University of Sydney. My core expertise includes enhancing large-scale 3D point cloud analysis, 3D mesh reconstruction, SLAM, and neural network-based segmentation models. Holding a Ph.D. from the University of Sydney with a focus on deep visual transfer learning, I excel in developing and optimizing deep learning algorithms for 2D and 3D visual tasks across both supervised and unsupervised learning paradigms.

Career Highlights

• 8+ years of AI and DL/ML research experience in 2D/3D Image/Video/Point Cloud multi-modal multi-domain supervised/unsupervised visual application processing system design.
• 3+ years of industry R&D experience at Bodymapp focusing on AWS MLOps pipeline design, 3D reconstruction, neural networks, measurement and recommendation system.
• 1 year of technical leading experience at Bodymapp, including method ideation and implementation, arranging meetings, brainstorming, retrospective and managing task priority.
• 1 year of tutoring and teaching experience at The University of Sydney, focusing on deep neural networks and transfer learning.

Research Interests

• 2D/3D Visual Applications
• Human Body Reconstruction and Tracking
• Model Generalization
• Deep Visual Transfer Learning
• Multi-sensor and Multi-domain Research

Featured Projects & Publications

Large-scale 3D Point Cloud Analysis

Large-scale 3D point cloud analysis for aerial multi-modal forestry data. Segment individual trees from large site plots. Use deep models to estimate tree parameters like height, diameter at breast height (DBH), volume, biomass, etc.

Precise 3D Mesh Reconstruction (Bodymapp)

Zero-to-One World-Class depth-only (protect privacy) mobile App for accurate scanning, reconstructing, and measuring the human body on various iOS devices (iPhones/iPads).

[US Patent 24] Zhang, W., et al. "Methods for generating a partial three-dimensional representation of a person." (in submission)

Cross-domain 3D Object Detection

Pioneering 3D point cloud based Cross-environment Object Detection in Autonomous Driving Scenarios.

[CVPR 21] Zhang, W., Li, W., & Xu, D. "SRDAN: Scale-aware and range-aware domain adaptation network for cross-dataset 3D object detection."

Multi-Modality Domain Adaptation

Progressive Modality Cooperation for Multi-Modality Domain Adaptation with missing modality generative model.

[T-IP 21] Zhang, W., Xu, D., Ouyang, W. & Zhang, J. "Progressive Modality Cooperation for Multi-Modality Domain Adaptation."

Cross-domain Cross-modality Recognition

Self-paced Collaborative and Adversarial Network with mutual multi-modal adaptability.

[T-PAMI 19] Zhang, W., Xu, D., Ouyang, W., & Li, W. "Self-paced collaborative and adversarial network for unsupervised domain adaptation."

Cross-domain Object Recognition

Collaborative and Adversarial Network for Unsupervised Domain Adaptation with pioneering use of adaptive pseudo labels and adversarial learning (GAN).

[US Patent 21] Zhang, W., et al. "Method for training deep neural network and apparatus." U.S. Patent Application No. 17/033,316.

[CVPR 18] Zhang, W., Ouyang, W., Li, W., & Xu, D. "Collaborative and adversarial network for unsupervised domain adaptation."

Cross-domain Model Compression

Model Compression using Progressive Channel Pruning which is proven effective combining transfer learning.

[T-CSVT 20] Guo, J., Zhang, W., Ouyang, W., & Xu, D. "Model Compression using Progressive Channel Pruning."

Cross-modality Pose Estimation

3D Hand Pose Estimation with Disentangled Cross-Modal Latent Space and modality generation.

[WACV 20] Gu, J., Wang, Z., Ouyang, W., Zhang, W., Li, J., & Zhuo, L. "3D Hand Pose Estimation with Disentangled Cross-Modal Latent Space."

Visitors