Embodied Minds Lab, Harvard University & Kempner Institute

less than 1 minute read

Published:

Visiting Undergraduate Research Assistant, Supervisor: Prof. Yilun Du & Dr. Ruojin Cai

Sep. 2025 - present

  • Developed an inverse generative modeling framework for scene understanding on both synthetic and real-world datasets, training a relation-conditioned composable diffusion model that generates scenes from structured object-attribute and spatial-relation inputs.
  • Focused on unsupervised object-relation discovery with Diffusion Models, enabling generative models to perform text-image attribution analysis on both synthesized and real-world images.
  • Investigated the potential of improving compositional generation with unsupervised relation discovery in a training-free manner, achieving SOTA performance on the 2D spatial relation dataset VISOR.