Frankenstein: Generating Semantic-Compositional 3D Scenes in One Tri-Plane DOI Open Access
Han Yan, Yang Li, Zhennan Wu

и другие.

Опубликована: Дек. 3, 2024

We present Frankenstein, a diffusion-based framework that can generate semantic-compositional 3D scenes in single pass. Unlike existing methods output single, unified shape, Frankenstein simultaneously generates multiple separated shapes, each corresponding to semantically meaningful part. The scene information is encoded one tri-plane tensor, from which Signed Distance Function (SDF) fields be decoded represent the compositional shapes. During training, an auto-encoder compresses tri-planes into latent space, and then denoising diffusion process employed approximate distribution of scenes. demonstrates promising results generating room interiors as well human avatars with automatically parts. generated facilitate many downstream applications, such part-wise re-texturing, object rearrangement or avatar cloth re-targeting.

Язык: Английский

Deep learning for 3D garment generation: A review DOI
Yuexin Sun, Zhenhua Hao, Zhaohui Wang

и другие.

Textile Research Journal, Год журнала: 2025, Номер unknown

Опубликована: Май 29, 2025

3D garment models enhance the consumer experience by enabling virtual trying-on and personalized customization. Additionally, they streamline design manufacturing processes, reduce resource waste, drive industry toward greater digitalization sustainability. Nevertheless, complexities of modeling have impeded its widespread adoption. Recent significant advances in deep learning catalyzed improvements model generation. This technology circumvents traditional time-consuming direct generation models, has garnered substantial attention. paper presents a comprehensive systematic review for It commences with an introduction to essential preliminaries, encompassing data representations, objectives tasks, generative datasets, evaluation methods. The categorizes works into three distinct areas: mesh, texture, pattern generation, providing in-depth analysis most recent advanced Furthermore, examines applications discusses current challenges, proposes directions future research, offering valuable insights continued exploration this rapidly expanding field.

Язык: Английский

Процитировано

0
Frankenstein: Generating Semantic-Compositional 3D Scenes in One Tri-Plane DOI Open Access
Han Yan, Yang Li, Zhennan Wu

и другие.

Опубликована: Дек. 3, 2024

We present Frankenstein, a diffusion-based framework that can generate semantic-compositional 3D scenes in single pass. Unlike existing methods output single, unified shape, Frankenstein simultaneously generates multiple separated shapes, each corresponding to semantically meaningful part. The scene information is encoded one tri-plane tensor, from which Signed Distance Function (SDF) fields be decoded represent the compositional shapes. During training, an auto-encoder compresses tri-planes into latent space, and then denoising diffusion process employed approximate distribution of scenes. demonstrates promising results generating room interiors as well human avatars with automatically parts. generated facilitate many downstream applications, such part-wise re-texturing, object rearrangement or avatar cloth re-targeting.

Язык: Английский

Процитировано

0