Generate Structured 3D Meshes from Single Images
PartCrafter is an innovative AI system that can convert single images into editable 3D mesh parts. Our technology understands object structure and generates high-quality 3D models.
Generation Time
0.5s
Editable Meshes
100%
F-Score
0.92
Quick Start
1. Clone Repository
git clone https://github.com/wgsxm/PartCrafter.git2. Install Dependencies
pip install -r requirements.txt3. Download Models
python scripts/download_models.py4. Run Generation
python generate.py --input image.jpg --output output/Technical Features
Compositional Latent Space
Our model uses an innovative compositional latent space representation to better understand object structure.
One-step Multi-mesh Output
Directly generate multiple editable 3D mesh parts from a single image, without additional post-processing steps.
High-quality Details
Generated 3D models feature fine geometric details and accurate texture mapping.
Editability
Generated 3D parts can be independently edited and combined, supporting flexible 3D modeling workflows.
Performance Metrics
| Model | F-Score | Generation Time | Editable |
|---|---|---|---|
| PartCrafter | 0.92 | 0.5s | Yes |
| Model A | 0.85 | 1.2s | No |
| Model B | 0.88 | 0.8s | No |
PartCrafter
F-Score
0.92
Generation Time
0.5s
Editable
Yes
Model A
F-Score
0.85
Generation Time
1.2s
Editable
No
Model B
F-Score
0.88
Generation Time
0.8s
Editable
No
demo.video
Research Paper
PartCrafter: Generating Structured 3D Meshes from Single Images
Our research paper details the technical principles and implementation methods of PartCrafter. The paper has been accepted by top conferences in computer vision and pattern recognition.