Garment Gallery
a belted floral-print linen mini dress
a black and white striped long sleeve shirt and dark denim jeans
a strapless ruffled plissé-chiffon tomato red gown
a white V-neck T-shirt and high-waisted plaid trousers
a green tie-dye sweatshirt and matching jogger pants
a floral-print short sleeve button-up shirt and white drawstring shorts
a color-blocked windbreaker jacket and black jogger pants
a green polo shirt and white linen pants
a white lace-up bodysuit and distressed boyfriend jeans
a white button-down shirt and black high-waisted trousers
a cutout floral-print satin-crepe mini dress
a striped long sleeve shirt and light wash mom jeans
a white oversized sweatshirt and athletic leggings
a man wearing a red plaid shirt and black skinny jeans
an oversized beige sweater, a chunky neck, and slim-fit trousers
a white slip and a yellow leather skirt
a black crew neck sweatshirt and grey jogger pants
a green velvet wrap dress with a knee-length hem
Abstract
High-fidelity 3D garment synthesis from text is desirable yet challenging for digital
avatar creation. Recent diffusion-based approaches via Score Distillation Sampling (SDS)
have enabled new possibilities but either intricately couple with human body or struggle
to reuse. We introduce ClotheDreamer, a 3D Gaussian-based method for generating wearable,
production-ready 3D garment assets from text prompts. We propose a novel representation
Disentangled Clothe Gaussian Splatting (DCGS) to enable separate optimization.
DCGS represents clothed avatar as one gaussian model but freezes body Gaussian splats.
To enhance quality and completeness, we incorporate bidirectional SDS to supervise
clothed avatar and garment RGBD renderings respectively with pose conditions and propose
a new pruning strategy for loose clothing. Our approach can also support custom clothing
templates as input. Benefiting from our design, the synthetic 3D garment can be easily
applied to virtual try-on and support physically accurate animation.
Extensive experiments showcase our superior and competitive performance.
Method
Given a text description, we first leverage ChatGPT to determine clothing ID types for initialization.
We introduce Disentangled Clothe Gaussian Splatting (DCGS), which represents clothed avatar as One-Gaussian model but
freezes body Gaussian splats to achieve separate supervision. With parsed Gaussian Splatting (GS) render,
we use Bidreactional SDS to guide clothing and body RGBD renderings separately with pose condition.
We also support template mesh input for versatile personalized 3D garment generation.
Animation Results
Auto Fitting
ClotheDreamer generated garments can fit different body shapes.
Related Works 🚀🚀
TexDreamer: Towards Zero-Shot High-Fidelity 3D Human Texture Generation.
GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models.
HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting.
DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation.
BibTeX
@misc{liu2024clothedreamer,
title={ClotheDreamer: Text-Guided Garment Generation with 3D Gaussians},
author={Yufei Liu and Junshu Tang and Chu Zheng and Shijie Zhang and Jinkun Hao and Junwei Zhu and Dongjin Huang},
year={2024},
eprint={2406.16815},
archivePrefix={arXiv},
primaryClass={cs.CV}}
Garment Gallery
a belted floral-print linen mini dress
a black and white striped long sleeve shirt and dark denim jeans
a strapless ruffled plissé-chiffon tomato red gown
a white V-neck T-shirt and high-waisted plaid trousers
a green tie-dye sweatshirt and matching jogger pants
a floral-print short sleeve button-up shirt and white drawstring shorts
a color-blocked windbreaker jacket and black jogger pants
a green polo shirt and white linen pants
a white lace-up bodysuit and distressed boyfriend jeans
a white button-down shirt and black high-waisted trousers
a cutout floral-print satin-crepe mini dress
a striped long sleeve shirt and light wash mom jeans
a white oversized sweatshirt and athletic leggings
a man wearing a red plaid shirt and black skinny jeans
an oversized beige sweater, a chunky neck, and slim-fit trousers
a white slip and a yellow leather skirt
a black crew neck sweatshirt and grey jogger pants
a green velvet wrap dress with a knee-length hem
Abstract
High-fidelity 3D garment synthesis from text is desirable yet challenging for digital avatar creation. Recent diffusion-based approaches via Score Distillation Sampling (SDS) have enabled new possibilities but either intricately couple with human body or struggle to reuse. We introduce ClotheDreamer, a 3D Gaussian-based method for generating wearable, production-ready 3D garment assets from text prompts. We propose a novel representation Disentangled Clothe Gaussian Splatting (DCGS) to enable separate optimization. DCGS represents clothed avatar as one gaussian model but freezes body Gaussian splats. To enhance quality and completeness, we incorporate bidirectional SDS to supervise clothed avatar and garment RGBD renderings respectively with pose conditions and propose a new pruning strategy for loose clothing. Our approach can also support custom clothing templates as input. Benefiting from our design, the synthetic 3D garment can be easily applied to virtual try-on and support physically accurate animation. Extensive experiments showcase our superior and competitive performance.
Method
Given a text description, we first leverage ChatGPT to determine clothing ID types for initialization. We introduce Disentangled Clothe Gaussian Splatting (DCGS), which represents clothed avatar as One-Gaussian model but freezes body Gaussian splats to achieve separate supervision. With parsed Gaussian Splatting (GS) render, we use Bidreactional SDS to guide clothing and body RGBD renderings separately with pose condition. We also support template mesh input for versatile personalized 3D garment generation.
Animation Results
Auto Fitting
ClotheDreamer generated garments can fit different body shapes.
Related Works 🚀🚀
GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models.
HumanGaussian: Text-Driven 3D Human Generation with Gaussian Splatting.
DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation.
BibTeX
@misc{liu2024clothedreamer,
title={ClotheDreamer: Text-Guided Garment Generation with 3D Gaussians},
author={Yufei Liu and Junshu Tang and Chu Zheng and Shijie Zhang and Jinkun Hao and Junwei Zhu and Dongjin Huang},
year={2024},
eprint={2406.16815},
archivePrefix={arXiv},
primaryClass={cs.CV}}