CV
Academic CV and selected research experience.
Contact Information
| Name | Xiaoqiang Shi |
| Professional Title | PhD Student |
| shixiaoqiang21@mails.ucas.ac.cn |
Professional Summary
PhD student in Computer Application Technology at the University of Chinese Academy of Sciences, working on multimodal large language models, agents, computer vision, and 3D Gaussian avatar reconstruction.
Education
-
2021 - Present China
PhD candidate
University of Chinese Academy of Sciences
Computer Application Technology
- Consecutive master-doctoral program.
- Research interests include multimodal LLMs, agents, computer vision, and 3D Gaussian avatar reconstruction.
-
2017 - 2021 Liaoning, China
Bachelor of Engineering
University of Science and Technology Liaoning
Mechanical Design, Manufacturing and Automation
Research Experience
Publications
-
2026 3DGA: 3D Avatar Animation from Monocular Video via Deformable Gaussian Splatting
Pattern Recognition
Co-first author work on monocular-video 3D avatar animation with deformable Gaussian Splatting. SCI Q1 Top, CCF B.
-
2023 BSSNet: A Real-Time Semantic Segmentation Network for Road Scenes Inspired from AutoEncoder
IEEE Transactions on Circuits and Systems for Video Technology
First-author work on real-time road-scene semantic segmentation with a focus on accuracy, speed, and model complexity. SCI Q1 Top, CCF B.
-
2026 CANVAS: Any-Shot Animatable 3D Gaussian Head Avatars from 1-K Images
Submitted to ACM Multimedia 2026
First-author work on any-shot animatable 3D Gaussian head avatars.
-
2026 HiFiAvatar: Prior-Guided Gaussian Appearance Learning for High-Fidelity Animatable Head Avatars from Monocular Video
Submitted to ACM Multimedia 2026
First-author work on high-fidelity animatable head avatars from monocular video.
-
2026 3DGA: Topology-Aware 3D Gaussian Head Avatars with Barycentric Parameterization and Layered Densification
Submitted to IEEE Transactions on Image Processing
First-author work on topology-aware 3D Gaussian head avatars. Submitted to SCI Q1 Top / CCF A venue.
Projects
-
Lightweight LLM Reproduction and Multimodal Extension
Reproduced a small-parameter decoder-only LLM pipeline from pretraining through SFT/LoRA, reinforcement learning, inference, and multimodal extension.
- Covered tokenizer, data construction, training, fine-tuning, RL training, sampling, KV cache, and deployment.
- Extended the pipeline toward speech-vision-language multimodal modeling.
-
Coding Agent Prototype
Built a code-oriented LLM agent prototype that closes the loop from user task understanding to file edits, command execution, verification, and error repair.
- Designed task parsing, context management, tool calling, and result feedback workflows.
- Supported code understanding, modification, validation, and iterative debugging.
-
3D Gaussian Human and Head Avatar Reconstruction
Developed a series of 3D Gaussian avatar methods for animatable human/head reconstruction and high-fidelity rendering.
- Worked on deformable Gaussian representations, topology constraints, barycentric parameterization, layered densification, and prior-guided appearance learning.
- Related work includes 3DGA, CANVAS, and HiFiAvatar.
-
BSSNet Real-Time Road Scene Semantic Segmentation
Designed a real-time semantic segmentation network for road scenes inspired by autoencoder structures.
- Balanced segmentation accuracy, inference speed, and model complexity.
- Published in IEEE TCSVT.