Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

Author ORCID Identifier



Open Access Dissertation

Document Type


Degree Name

Doctor of Philosophy (PhD)

Degree Program

Computer Science

Year Degree Awarded


Month Degree Awarded


First Advisor

Evangelos Kalogerakis

Second Advisor

Rui Wang

Third Advisor

Subhransu Maji

Fourth Advisor

Benjamin S Jones

Subject Categories

Graphics and Human Computer Interfaces


In this dissertation I will investigate algorithms that analyze stylistic properties of 3D shapes and automatically synthesize shapes given style specifications. I will start by introducing a structure-transcending method for style similarity evaluation between 3D shapes. Inspired by observations about style similarity in art history literature, we propose an algorithmically computed style similarity measure which identifies style related elements on the analyzed models and collates element-level geometric similarity measurements into an object-level style measure consistent with human perception. To achieve this consistency we employ crowdsourcing to learn the relative perceptual importance of a range of elementary shape distances and other parameters used in our measurement from participant answers to cross-structure style similarity queries. I will then describe an algorithm that utilizes this learned style similarity measure to synthesize 3D models of man-made shapes. The algorithm combines user-specified style, described via an exemplar shape, and functionality, encoded by a functionally different target shape. We transfer the exemplar style to the target via a sequence of compatible element-level operations where the compatibility is a learned metric that estimates the impact of each operation on the edited shape. We use this metric to cast style transfer as a tabu search, which incrementally updates the target shape using compatible operations, progressively increasing its style similarity to the exemplar while strictly maintaining its functionality at each step. Finally I will propose a method for reconstructing 3D shapes following style aspects of given 2D drawings. Our method takes line drawings as input and converts them into surface depth and normal maps from several output viewpoints via a deep convolutional neural network with multi-view encoder-decoder architecture. The multi-view maps are then consolidated into a dense coherent 3D point cloud by solving an optimization problem that fuses depth and normal information across all output viewpoints. The output point cloud is then converted into a polygon mesh representation, which is further fine-tuned to match the input sketch more precisely.