Conformal Cache: Reliable Proxy-Discrepancy Caching for Fast Generative Inference
Computer Vision · Multimodal Learning
Chengming Xu
Senior Researcher at Youtu Lab, Tencent · Ph.D. in Data Science, Fudan University
I work on deep learning for computer vision with limited supervision, with recent interests in visual in-context learning, multimodal reasoning, and controllable generation.
Research interests: few-shot learning, visual in-context learning, vision-language models, and video generation/editing.
Selected Recent Works
View full list →* Equal contribution / core contributor. † Corresponding author.
Evidence-RL: Towards Evidence-intensive Visual Reasoning
PixVerve: Advancing Native UHR Image Generation to 100MP with a Large-Scale High-Quality Dataset
What Semantics Survive the Connector? Diagnosing VLM-to-DiT Alignment in Video Editing
FFP-300K: Scaling First-Frame Propagation for Generalizable Video Editing