Zhiwen Chen (陈志文)

I'm currently working as a Staff Algorithm Engineer at Alibaba Group where I'm leading the PixelAI Algorithm Team of Tao Technology. Previously before 2017, I worked as a Video Analytic Researcher at Trakomatic Pte. Ltd., Singapore for several years.

I received the B.E. degree in computer science from SJTU, under the supervision of Prof. Fan Wu in 2012 and the M.E. degree in computer science from NUS in 2014.

Email  /  LinkedIn  /  CV

I'm interested in Computer Vision, in particular, Human Reconstruction, Animatable Avatar, Human-Computer Interaction, etc. Below are some highlighted publications.

We introduce TaoAvatar, which generates photorealistic, topology-consistent 3D full-body avatars from multi-view sequences. It provides high-quality, real-time rendering with low storage requirements, compatible across various mobile and AR devices like the Apple Vision Pro.

We introduce SaMer, a fine-grained, scenario-adaptive evaluator that dynamically adjusts evaluation dimensions based on query context.

We present GaussianTalker, a novel method for audio-driven talking head synthesis based on 3D Gaussian Splatting. It outperforms existing state-of-the-art methods in talking head synthesis, delivering precise lip synchronization and exceptional visual quality. It also achieves rendering speeds of 130 FPS on NVIDIA RTX4090 GPU.

We present a novel framework for 6DoF face pose estimation, where 2D features extracted from images and 3D features representing 3D shape interact with each other in a transformer architecture to learn the 2D-3D correspondence.

We present 3D Avatar Reconstruction in the wild (ARwild), which first reconstructs the implicit skinning fields in a multi-level manner.

We proposed an attention-based model called Context Attention Network (CANet), which integrates the context extraction module in a UNet architecture, can effectively improve the network’s ability to extract the skeleton pixels. We were evaluated 1st place on CVPR DLGC Workshop and Challenge.

We established a simple but effective baseline for single human motion forecasting without visual and social information. We were evaluated 1st place on ICCV SoMoF Workshop and Challenge.

These include workshops, challenges and awards.

We contribute a large-scale dataset, MVP-Human (Multi-View and Multi-Pose 3D Human), which contains 250 subjects. Each subject has 15 type of different poses. Each pose contains 8-view RGB images.

Template copied from Jon Barron and Matiur Rahman Minar.