CoMA: Compositional Human Motion Generation with Multi-modal Agents
A multi-modal, compositional framework that refines complex human motion generation from textual descriptions, producing more realistic and diverse motion sequences.
Baltimore, Maryland
About me
Shenghan Zhou
MSE Student in Computer Science at Johns Hopkins University
I work at the intersection of computer vision, generative models, and 3D understanding. I’m currently interested in vision-language-action models (VLA) and embodied AI—building intelligent agents that can perceive, reason, and act in the physical world.
Research interests
- VLA
- Embodied AI
- Computer Vision
- Generative AI
01 / Selected work
Publications
Research on controllable generation and efficient visual models.
Accelerating Style Transfer: Enhancing Efficiency of Diffusion-Based Models with Advanced Sampling Methods
An investigation of accelerated sampling for diffusion-based style transfer. UniPC achieved a fivefold speedup for high-quality stylized image generation in the study.
02 / Background
Education
2025 — 2027
Current
Johns Hopkins University
M.S.E. in Computer Science
Graduate study in computer science with interests in visual computing and generative artificial intelligence.
2024
University of California, Berkeley
Berkeley Global Access Program
Coursework in computer vision, artificial intelligence, and computer networks. View CS 180 projects
Summer 2024
University of California, Irvine
UCInspire Research Program · GPA 4.0/4.0
Worked with Prof. Xiaohui Xie on human motion generation.
2021 — 2025
Chongqing University
B.Eng. in Artificial Intelligence
GPA 92/100 (3.848/4.0) · Major rank 1/63 · Grade rank 3/253
03 / Experience
Selected projects
Applied work spanning recognition, graph learning, detection, and 3D vision.
01
Graduation project · 2024
Single-Image 3D Human Reconstruction
A diffusion-based reconstruction pipeline using SMPL-X priors, multi-view synthesis, ViT encoders, and cross-view feature fusion to recover detailed geometry and texture.
- Diffusion
- SMPL-X
- ViT
- 3D Vision
02
Team lead · 2023–2024
Object Detection in Adverse Weather
Adapted IA-YOLO with YOLOv3 and a channel-attention mechanism for robust detection under challenging weather conditions.
- PyTorch
- YOLOv3
- Attention
03
Research project · 2023–2024
Dynamic Graph Fraud Detection
Combined spatial and temporal aggregation to capture evolving patterns in financial transaction graphs.
- GNN
- Temporal ML
- Fraud Detection
04
Independent study · 2023
AI Face Recognition System
Built an end-to-end facial authentication system with OpenCV, TensorFlow, and Flask, reaching 91% test accuracy.
- TensorFlow
- OpenCV
- Flask
04 / Recognition
Awards &
honors
- Outstanding Graduate of Chongqing City Top 1%
- Honorable Mention, Mathematical Contest in Modeling
- First Prize, CUMCM Top 1%
- National Scholarship Top 0.2%
- First Prize, CQU Mathematics Contest Top 10%
Let’s connect