Baifeng Shi

I am a Ph.D. student advised by Prof. Trevor Darrell at UC Berkeley. Previously, I graduated from Peking University with a B.S. degree in computer science.

I build generalist vision and robotic models.

Email  /  Google Scholar  /  Github  /  CV  /  WeChat

We find that smaller vision models (e.g., ViT-B or Vit-L) run on multiple image scales are usually better than larger models (e.g., ViT-H, ViT-G).

We formulate humanoid locomotion as a next token prediction problem. This enables learning to walk from in-the-wild data such as Youtube videos.

We make imitation learning easier by MAE pre-training on sensorimotor sequences.

We find that previous transfer learning methods (e.g., fine-tuning, LoRA, prompt tuning) fail to focus the model's attention on the features relevant to the downstream tasks. We show that refocusing the model's attention on task-relevant features by top-down attention can largely improve the downstream performances.

We build ViTs with the ability of top-down attention, i.e., steering its attention to specific objects when given a prompt.

Last updated: Mar 18, 2024