PI: Joe Sventek

The goal of this projects is to prototype to segment 3D objects into equivalence classes using known Deep Learning techniques; understand performance of known techniques.

Methods

• Focus on 2D and 3D computer vision images of objects
• Define labelling format for these images
• Understand existing SIFT-based algorithm accuracy and performance
• Apply known Deep Learning techniques, comparing accuracy and performance with SIFT-based techniques
• Produce proof-of-concept Python implementation
• Particularly focus on any obvious patterns exhibited by false positive/negative classifications