Xuming He

Introducing a new method for more accurate scene graph generation.

A new training method enhances vision-language models' performance in zero-shot tasks.

A new framework transforms image interpretation through open-vocabulary scene graphs.

SPOT framework improves clustering effectiveness in imbalanced datasets using pseudo-labels.

A framework to assess LLMs' abilities in data-related tasks with code interpreters.

Developing adaptive methods for 3D data segmentation to identify new object classes.

A new method enhances language model training using self-generated feedback.

FastGrasp improves how we simulate human-like grasping with efficiency and realism.