Dissertation Defense

Event: 3D Object Representations for Recognition

Yu Xiang

Object recognition from images is a longstanding and challenging problem in computer vision. The main challenge is that the appearance of objects in images is affected by a number of factors, such as illumination, scale, camera viewpoint, intra-class variability, occlusion, truncation, and so on. How to handle all these factors in object recognition is still an open problem. In this dissertation, I present my efforts in building 3D object representations for object recognition. Compared to 2D appearance based object representations, 3D object representations can capture the 3D nature of objects and better handle viewpoint variation, occlusion and truncation in object recognition.

I introduce three new 3D object representations: the 3D aspect part representation, the 3D aspectlet representation and the 3D voxel pattern representation. These representations are built to handle different challenging factors in object recognition. The 3D aspect part representation is able to capture the appearance change of object categories due to viewpoint transformation. The 3D aspectlet representation and the 3D voxel pattern representation are designed to handle occlusions between objects in addition to viewpoint change. Based on these representations, we propose new object recognition methods and conduct experiments on benchmark datasets to verify the advantages of our methods.

Furthermore, we introduce, PASCAL3D+, a new large scale dataset for 3D object recognition by aligning objects in images with 3D CAD models. We also propose two novel methods to tackle object co-detection and multiview object tracking using our 3D aspect part representation, and a novel Convolutional Neural Network-based approach for object detection using our 3D voxel pattern representation. In order to track multiple objects in videos, we introduce a new online multi-object tracking framework based on Markov Decision Processes. Lastly, I conclude the dissertation and discuss future steps for 3D object recognition.

Sponsored by


Faculty Host

Silvio Savarese and Alfred Hero