Abstract:Geometric modeling is a fundamental problem in computer graphics. The continuous growth of 3D models in public repositories has shifted research focus from computing local, low-level geometry features such as curvature and textures to high-level semantic information such as shape parts information and shape structural characteristics (symmetry, parallelism, etc.). However, automatically computing such semantic information is essentially an ill-posed problem due to the ambiguities in the definition of shape semantics. This thesis focuses on developing interactive approaches to aid the shape analysis process. We leverage user assistance to exploit shape semantics, and to respect and preserve them during manipulation. In particular, this thesis aims at advancing state-of-the-art interactive techniques in two specific geometric applications: shape segmentation and shape manipulation. First, we introduce two interactive tools for shape segmentation, which we call cross-boundary brushes and dot scissor. Both tools offer very simple and easy-to-use user interfaces that operate at interactive rates. In contrast to existing state-of-the-art interactive segmentation tools, our tools allow the user to cut out meaningful and functional components in most cases using only a single mouse stroke or click near boundary regions, making them very convenient to use. We adopt the concept of isolines of harmonic fields as cutting boundaries in designing both tools. We show that the propagation properties and the differentiating power of the harmonic fields allow effective computation of shape semantic boundaries for segmentation purposes. Second, we developed an editing framework that first extracts the shape's structural features and preserves them during user manipulation. In contrast to traditional shape editing frameworks, the system operates at the component level and takes a shape's structural characteristics such as inter-relations among semantic components as modeling constraints, enabling an effective structure-preserving editing tool. We show that user assistance is essential in accurately revealing complex shape structures. We use a semi-automatic shape segmentation process as a prior step to facilitate the analysis of shape structures and inter-relations, and show that these shape analysis results play an important role in preserving a shape's global features during user manipulation.

Deformable Geometry Based Semantic Reconstruction from Scene Graphs.

DeformSg2im: Scene Graph Based Multi-Instance Image Generation with a Deformable Geometric Layout.

Semantic-based Interactive Shape Analysis and Manipulation

Scene Graph Disentanglement and Composition for Generalizable Complex Image Generation

Semantic Image Manipulation Using Scene Graphs

Deformable Generator Networks: Unsupervised Disentanglement of Appearance and Geometry

SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction with Geometry Constrain

Learning Canonical Representations for Scene Graph to Image Generation

Image Generation from Scene Graph with Object Edges

Semantic Reconstruction based on RGB Image and Sparse Depth

“Edutainment 2017” a Visual and Semantic Representation of 3D Face Model for Reshaping Face in Images

Enhanced 3D Shape Reconstruction With Knowledge Graph of Category Concept

Scene Graph Generation With External Knowledge and Image Reconstruction

Exploiting Relationship for Complex-scene Image Generation

A Unified Framework for Piecewise Semantic Reconstruction in Dynamic Scenes Via Exploiting Superpixel Relations.

Text Pared into Scene Graph for Diverse Image Generation.

3D Scene Graph: A Structure for Unified Semantics, 3D Space, and Camera

Semantic-guided 3D building reconstruction from triangle meshes

From sketch to reality: precision-friendly 3D generation technology

Scene Graph Generation: A Comprehensive Survey

Topology-aware Human Avatars with Semantically-guided Gaussian Splatting