Repolode
Submit
hi3d-official

hi3d-official/Hi3Dv1.0

Open-source inference code and model weights that refine a coarse 3D mesh into high-resolution geometry, guided by a reference image.

View on GitHubWebsite
Stars
238
Forks
17
Open issues
0
Last commit
2 days ago

What is it?

Hi3D v1.0 takes a coarse mesh and a reference image of the same object and produces a refined, high-resolution mesh. It is built on the Sparc3D sparse representation: the pipeline progressively samples sparse latents, decodes them with a sparse VAE, and extracts a mesh, repeating the process over up to three passes.

The pipeline supports 512, 1024 and 1536 output resolutions, with 1536 running all three passes. The output is geometry only, with no textures or materials. Python users load the model from Hugging Face through the Sparc3DPipeline class, and a command-line script refines an existing mesh without any dependency on a coarse generator.

Inference requires Linux, Python 3.10 and an NVIDIA GPU with compute capability 8.0 or newer, and there is no CPU path. The repository does not include training code or evaluation tooling.

Key features

  • 01Refines a coarse mesh using a reference image of the same object.
  • 02Outputs high-resolution geometry at 512, 1024 or 1536 resolution.
  • 03Loads model weights directly from Hugging Face or a local directory.
  • 04Provides a Python API and a command-line script for mesh refinement.
  • 05Supports sequential CPU offload and per-step progress callbacks.

Use cases

  • Adding finer surface detail to meshes from a 3D generator.
  • Refining low-poly or coarse models using a reference photo.
  • Batch-processing existing meshes in a 3D asset pipeline.

Quick start

  1. 1

    Clone the repository

    git clone https://github.com/hi3d-official/Hi3Dv1.0.git
  2. 2

    Enter the project directory

    cd Hi3Dv1.0
  3. 3

    Install PyTorch with CUDA 12.4 (then install flash-attn as described in the README)

    pip install torch==2.5.0+cu124 torchvision==0.20.0+cu124 --extra-index-url https://download.pytorch.org/whl/cu124
  4. 4

    Install the project from source with the inference extra

    pip install -c docker/constraints.txt -e ".[inference]"

Things to know

  • Requires Linux and an NVIDIA GPU with compute capability 8.0 or newer; there is no CPU inference path.
  • The 1536 resolution has only been validated on a 96 GB NVIDIA H20, so smaller cards are unverified.
  • Commercial use is allowed only under the custom Hi3D Open License, which has user-count and revenue thresholds and branding requirements.
  • Installation is from source only, with no PyPI release, and needs a matching CUDA toolkit and flash-attn build.

Topics

#3d-generation#geometry-refinement#image-to-3d#sparse-representation#pytorch#mesh#sparc3d

FAQ

What is Hi3Dv1.0?
Open-source inference code and model weights that refine a coarse 3D mesh into high-resolution geometry, guided by a reference image. Hi3D v1.0 takes a coarse mesh and a reference image of the same object and produces a refined, high-resolution mesh. It is built on the Sparc3D sparse representation: the pipeline progressively samples sparse latents, decodes them with a sparse VAE, and extracts a mesh, repeating the process over up to three passes.
Is Hi3Dv1.0 free to use? What license does it have?
Hi3Dv1.0 is open source under the Other license. Check the license terms before using it in commercial products.
Is Hi3Dv1.0 still maintained?
Actively maintained. The last commit to Hi3Dv1.0 was on Oct 23, 2026.
What can I use Hi3Dv1.0 for?
Adding finer surface detail to meshes from a 3D generator. Refining low-poly or coarse models using a reference photo. Batch-processing existing meshes in a 3D asset pipeline.
How do I install Hi3Dv1.0?
Clone the repository: git clone https://github.com/hi3d-official/Hi3Dv1.0.git Enter the project directory: cd Hi3Dv1.0 Install PyTorch with CUDA 12.4 (then install flash-attn as described in the README): pip install torch==2.5.0+cu124 torchvision==0.20.0+cu124 --extra-index-url https://download.pytorch.org/whl/cu124 Install the project from source with the inference extra: pip install -c docker/constraints.txt -e ".[inference]"
Where can I get Hi3Dv1.0?
The source code and installation instructions are on GitHub: https://github.com/hi3d-official/Hi3Dv1.0

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects