Yang Fei (费阳)
I am a senior undergraduate at The Hong Kong University of Science and Technology (HKUST), majoring in Computer Science and Mathematics. I also spent time as an exchange student at the University of Washington (UW).
I have been fortunate to conduct research under the supervision of Prof. Qifeng Chen at HKUST and Prof. Ranjay Krishna at UW. My research centers on Computer Vision, with a particular focus on enhancing motion modeling in generative video models. I aim to build motion-centric world models where temporal consistency is prioritized, pushing video generation toward a digital-twin-like reality.
News
- [2025.12] Our new preprint, "Structure from Tracking" (first author), is now available! See the project page.
- [2025.06] VideoVAE+ (co-first author) has been accepted to ICCV 2025! Code and weights are available here.
- [2025.04] Received the The Hong Kong, China - Asia-Pacific Scholarship!
- [2025.01] Started an exchange program at the University of Washington (Seattle).
Research
Building on my experience improving motion architectures, from the tokenizer to the diffusion stage, I am focused on advancing motion modeling in video generation.
Interactive Causal Motion: I aim to transform video generation models into interactive simulators that support active intervention rather than mere observation, creating environments where agents can effectively plan and act.
Geometrically Consistent Motion: I am focused on training efficiency and geometric fidelity. Specifically, I investigate methods to address the sample inefficiency of learning 3D consistency solely from 2D statistics.
Physically Grounded Motion: My goal is to bridge the gap between visual plausibility and physical correctness. I explore how to enforce fundamental constraints to handle complex physical phenomena often underrepresented in training data.