Kaican Li

InSight-o3: Empowering Multimodal Foundation Models with Generalized Visual Search

CannyEdit: Selective Canny Control and Dual-Prompt Guidance for Training-free Image Editing

Dual Risk Minimization: Towards Next-Level Robustness in Fine-tuning Zero-Shot Models

Consistency Regularization for Domain Generalization with Logit Attribution Matching

Robustness May be More Brittle than We Think under Different Degrees of Distribution Shifts