Jingfeng Yang

I am a junior undergraduate at UC Berkeley, majoring in Computer Science. I am a member of BAIR (Berkeley Artificial Intelligence Research), where I have the privilege of working under the guidance of Prof. Yi Ma and Prof. Trevor Darrell.

In my spare time, I enjoy traveling and jogging. I also love learning new languages and can speak Mandarin, English, and German, with a little bit of Korean, Japanese, and Spanish.

Email  /  CV  /  Scholar  /  Github  /  Hugging Face  /  Linkedin

Research

My primary research interests focus on interpreting the process of learning visual representations — specifically, how rich visual semantics are compressed into compact embeddings using self-supervised or minimally supervised approaches — and evaluating how these representations perform in a variety of downstream tasks.

[Project Page] [Preprint] [PDF] [Github]

We question a core assumption held by dominant language-image alignment approaches like CLIP — that text and vision encoders should be jointly trained from scratch to achieve optimal alignment. We present LIFT, which pre-computes fixed text representations from LLMs and solely trains the vision encoder.

[Project Page] [Preprint] [PDF] [Github]

We present Unsupervised SAM (UnSAM), a segment anything model for interactive and automatic whole-image segmentation, trained entirely on pseudo masks generated by our unique divide-and-conquer pipeline without any human annotations.

Other Projects

[Project Page] [PDF] [Github]

We propose DREAMoR, a diffusion-based motion prior framework that reconstructs physically plausible human motion from sequences corrupted by occlusions, poor visibility, and ambiguous poses in monocular motion capture.

Teaching

Spring 2025, by Prof. Thomas Courtade

Spring 2024, by Prof. Gireeja Ranade

Fall 2023, by Swupnil Sahai and Raza Khan.

Spring 2023, by Prof. Joseph Gonzalez and Swupnil Sahai

This website is built on this public repo. Do not scrape the HTML from this page itself.