Edit3D-Bench

Paper | Project Page | Code

Edit3D-Bench is a benchmark for 3D editing evaluation, introduced in the paper VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space. This dataset comprises 100 high-quality 3D models, with 50 selected from Google Scanned Objects (GSO) and 50 from PartObjaverse-Tiny. For each model, we provide 3 distinct editing prompts. Each prompt is accompanied by a complete set of annotated 3D assets, including

• original 3D asset with rendered images
• 3D mask specifying the editing region with rendered images
• 2D mask of the edit region
• 2D edited image generated by FLUX.1 Fill

Preview

Explore our dataset in Project Page.

Which tasks will benefit from our dataset?

• 3D Editing

⚙️ Getting Started

Download the Dataset

To download the full dataset, you can use the following code. If you encounter any issues, please refer to the official Hugging Face documentation.

# Make sure you have git-lfs installed (https://git-lfs.com)
git lfs install

# When prompted for a password, use an access token with write permissions.
# Generate one from your settings: https://huggingface.co/settings/tokens
git clone https://huggingface.co/datasets/huanngzh/Edit3D-Bench

# If you want to clone without large files - just their pointers
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/datasets/huanngzh/Edit3D-Bench

Dataset Structure

data/
├── metadata.json                   # Dataset metadata file (dataset, object_name, prompt)
├── GSO/                            # Google Scanned Objects dataset
│   ├── [object_name]/
│   │   ├── source_model/           # Original 3D model
│   │   │   ├── model.glb             # Original 3D model file (GLB format)
│   │   │   ├── render/               # Original model rendered images
│   │   │   ├── video_rgb.mp4         # Original model RGB video
│   │   │   ├── video_normal.mp4      # Original model normal video
│   │   │   └── video_mask.mp4        # Original model mask video
│   │   ├── prompt_1/               # Annotation for prompt 1
│   │   │   ├── 2d_edit.png           # 2D edited image
│   │   │   ├── 2d_mask.png           # 2D mask image for editing
│   │   │   ├── 2d_render.png         # 2D render image of original model
│   │   │   ├── 2d_visual.png         # 2D visualization image
│   │   │   ├── 3d_edit_region.glb    # 3D edit region model
│   │   │   └── render/               # Rendered images of 3D mask
│   │   ├── prompt_2/               # Annotation for prompt 2
│   │   └── prompt_3/               # Annotation for prompt 3
│   └── ...
└── PartObjaverse-Tiny/             # PartObjaverse-Tiny dataset
    ├── [object_id]/
    │   ├── source_model/           # Original 3D model
    │   ├── prompt_1/               # Annotation for prompt 1
    │   ├── prompt_2/               # Annotation for prompt 2
    │   └── prompt_3/               # Annotation for prompt 3
    └── ...

Evaluation

Check details in our github repo.

🧷 Citation

@article{li2025voxhammer,
  title = {VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space},
  author = {Li, Lin and Huang, Zehuan and Feng, Haoran and Zhuang, Gengxiong and Chen, Rui and Guo, Chunchao and Sheng, Lu},
  journal = {arXiv preprint arXiv:2508.19247},
  year = {2025},
  url = {https://huggingface.co/papers/2508.19247}
}