Hi👋3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging

1The Chinese University of Hong Kong, Shenzhen,   2ByteDance,   3Tsinghua University
teaser image

Hi👋3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging.

Abstract

With the growing demand for high-fidelity 3D models from 2D images, existing methods still face significant challenges in accurately reproducing fine-grained geometric details due to limitations in domain gaps and inherent ambiguities in RGB images. To address these issues, we propose Hi3DGen, a novel framework for generating high-fidelity 3D geometry from images via normal bridging. Hi3DGen consists of three key components: (1) an image-to-normal estimator that decouples the low-high frequency image pattern with noise injection and dual-stream training to achieve generalizable, stable, and sharp estimation; (2) a normal-to-geometry learning approach that uses normal-regularized latent diffusion learning to enhance 3D geometry generation fidelity; and (3) a 3D data synthesis pipeline that constructs a high-quality dataset to support training. Extensive experiments demonstrate the effectiveness and superiority of our framework in generating rich geometric details, outperforming state-of-the-art methods in terms of fidelity. Our work provides a new direction for high-fidelity 3D geometry generation from images by leveraging normal maps as an intermediate representation.

teaser image

Method

The First Stage: Image-to-Normal Estimation. Left part: Illustration of Noise-injected Regressive Normal Estimation (NiRNE); Right part: Noisy label at high-frequency regions in real-domain data.

Left Top Image

Normal estimation results. Qualitative comparisons.

Left Bottom Image

The Second stage: Normal-to-Geometry Generation. An illustration of Normal-Regularized Latent Diffusion (NoRLD).

Right Top Image

The construction procedure of the proposed DetailVerse dataset, which contains high-quality synthesized 3D assets to support the training of Hi3DGen. The pipeline contains three steps: text prompt coleection, image generation, and 3D assests synthesis.

Right Bottom Image

Image-to-3D Generation Results

Click on the images below to see our generated results as geometry-only meshes in a 3D viewer.

💡Tips

● Scroll to zoom in/out

● Drag to rotate

● Press "shift" and drag to pan

Comparison to Other Methods

Select a method from the dropdown menu to compare the results of Hi3DGen with it side by side.

Hi3DGen
💡Tips

● Scroll to zoom in/out

● Drag to rotate

● Press "shift" and drag to pan

*Geometry-only results.

Generation Gallery

Zoom in for better visualization of each 3D model.

BibTeX

@article{ye2025hi3dgen,
  title={Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridging},
  author={Ye, Chongjie and Wu, Yushuang and Lu, Ziteng and Chang, Jiahao and Guo, Xiaoyang and Zhou, Jiaqing and Zhao, Hao and Han, Xiaoguang},
  journal={arXiv preprint arXiv:2503.22236}, 
  year={2025}
}