Skip to Main content Skip to Navigation
Theses

Reverse-Engineering Fashion Products : From a single-view Sketch to a 3D Model

Amélie Fondevilla 1
1 IMAGINE [2016-2019] - Intuitive Modeling and Animation for Interactive Graphics & Narrative Environments [2016-2019]
Inria Grenoble - Rhône-Alpes, Grenoble INP [2007-2019] - Institut polytechnique de Grenoble - Grenoble Institute of Technology [2007-2019], LJK - Laboratoire Jean Kuntzmann
Abstract : The efficient creation of virtual 3D content is a major issue in Computer Graphics industry and research. This thesis addresses the modeling of sewed 3D objects such as clothes, shoes or accessories from 2D sketched inputs. While existing related approaches are either based on multi-view inputs or on character dependent interfaces, we propose in this thesis to start from an annotated single-view input in order to create 3D content satisfying geometric constraints that are specific to fabric, such as developability, or fold appearance and distribution. Our goal is to exploit the expressiveness user-drawn sketches to guide the modeling of plausible objects, using a priori geometric knowledge.We first present a single-view reconstruction approach for symmetric piece-wise developable objects from an annotated photo. Using the assumption of mirror-symmetry along with properties of smooth developable surfaces, we propose a system able to lift in 3 the silhouette, borders and seams drawn on the picture and reconstruct the surface of the object. Our method also provides topology information on the object, allowing the computation of 2D patterns for each piece of developable surface, which is necessary for manufacturing.While most of existing sketch-based modeling methods for garments aim at reconstructing border lines drawn on top of a view of the 3D virtual mannequin to be dressed, our approach uses a single 2D sketch as input. Our method first extracts generic features relative to the style of the sketched garment expressed as garment proportions, tight areas, silhouette shapes and folds. In particular, we propose a dedicated approach to extract robustly folds characteristics, even in the case of garment with deep folds leading to partially occluded garment borders. In a second step, we synthesize on a target virtual character a 3D garment surface exhibiting similar style, while adapting to the target pose and morphology that may be drastically different from the one depicted the sketch.
Complete list of metadatas

Cited literature [118 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02908437
Contributor : Abes Star :  Contact
Submitted on : Wednesday, July 29, 2020 - 9:27:11 AM
Last modification on : Wednesday, October 14, 2020 - 4:17:05 AM

File

FONDEVILLA_2019_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02908437, version 1

Collections

Citation

Amélie Fondevilla. Reverse-Engineering Fashion Products : From a single-view Sketch to a 3D Model. Artificial Intelligence [cs.AI]. Université Grenoble Alpes, 2019. English. ⟨NNT : 2019GREAM070⟩. ⟨tel-02908437⟩

Share

Metrics

Record views

146

Files downloads

85