Data-Driven Machine Learning in Computer Vision

We are interested in cutting-edge deep learning technologies to study various computer vision problems in bio-medicine, including but not limited to:

  • Learning features from 2D/3D imaging data
  • Learning shape descriptors from 3D models (point clouds, meshes, etc.)
  • Learning motion (deformation) characteristics from 3D dynamic models 

Computational Tools for 3D SEM/TEM

In this project, we develop a number of algorithms and software tools to reconstruct 3D models from a series of scanning or transmission electron microscopy images:

  • GPU-accelerated 3D surface reconstruction from 2D SEM imaging data
  • 3D surface characteristics analysis of SEM
  • Automated particle detection from 2D cryo-EM images
  • 3D atomic modeling from reconstructed cryo-EM volumes of molecular complexes 

Image Processing and Analysis

We are interested in a broad range of image processing and analysis problems. Of our particular interest are:

  • Image interpolation and registration on serial sectioning data
  • Mesh generation from 3D images
  • Image segmentation (graph-based, level sets, and deep learning methods)
  • 3D surface reconstruction from multi-view 2D images 

Geometric (Mesh) Modeling and Processing

This project involves a number of mesh generation and processing problems:

  • Feature-preserving surface mesh smoothing (reducing mesh bumpiness)
  • Quality-guaranteed surface mesh smoothing (improving angle distribution)
  • High-quality tetrahedral mesh generation from surface meshes
  • 3D shape descriptor and shape retrieval 

Molecular Shape Modeling

In this project we aim to explore the following specific problems:

  • High-quality surface and volumetric mesh generation from molecules
  • Effective geometric measurements and analysis of molecular surfaces
  • Non-rigid molecular shape matching 

Scientific Computing and Visualization

We are interested in scientific computing and 3D visualization, including:

  • Radial basis functions in interpolation and eigenanalysis of 3D point clouds
  • GPU-based parallel computing in biomedical simulations
  • Efficient visualization of 3D images and simulated data