File size: 3,762 Bytes
0a1186d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 | # InteriorFusion: Single Image → High-Quality Editable 3D Interior Scenes
**InteriorFusion** is a production-grade, open-source AI system for converting a single 2D interior photograph into a high-quality, editable 3D scene with textured meshes, Gaussian splats, and PBR materials.
## 🏆 Why InteriorFusion?
Current image-to-3D pipelines (TRELLIS, Hunyuan3D-2, TripoSR, Stable Fast 3D) are designed for **single objects**, not **rooms**. InteriorFusion is the first system specifically architected for **interior design and room reconstruction**.
### Key Differentiators
- **Scene-aware**: Understands room layouts, wall/floor topology, and spatial relationships
- **Multi-object coherent**: Furniture doesn't float; objects respect real-world scale
- **Editable**: Individual furniture objects are separable, movable, replaceable
- **PBR materials**: Metallic, roughness, normal maps for photorealistic rendering
- **Multi-format export**: GLB, FBX, OBJ, USDZ, Gaussian Splatting
- **Under 30s generation** on RTX 4090
## Architecture Overview
```
Single Interior Image
│
▼
┌─────────────────────────────────────┐
│ Phase 1: Scene Understanding │
│ - Depth Anything V2 (metric indoor) │
│ - Room layout estimation │
│ - Semantic segmentation │
└─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Phase 2: Multi-View Generation │
│ - Zero123++ / SyncDreamer │
│ - Depth-conditioned view synthesis │
└─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Phase 3: 3D Reconstruction │
│ - Room shell (walls/floor/ceiling) │
│ - Per-object Gaussian reconstruction │
│ - TRELLIS.2 for furniture detail │
└─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Phase 4: Scene Assembly │
│ - SpatialLM layout optimization │
│ - Collision detection & correction │
│ - Scale normalization │
└─────────────────────────────────────┘
│
▼
┌─────────────────────────────────────┐
│ Phase 5: Material & Texture │
│ - PBR material generation │
│ - Texture baking │
│ - Light estimation │
└─────────────────────────────────────┘
│
▼
Export: GLB / FBX / OBJ / USDZ / 3DGS
```
## Model Zoo
| Model | HF Path | Purpose | Size |
|-------|---------|---------|------|
| InteriorFusion-S | `stevee00/InteriorFusion-S` | Fast preview, 512³ | 1.5B |
| InteriorFusion-L | `stevee00/InteriorFusion-L` | Production quality, 1024³ | 4B |
| InteriorFusion-XL | `stevee00/InteriorFusion-XL` | Research quality, 2048³ | 10B |
## Quick Start
```bash
pip install -e .
python -m interiorfusion.infer --image room_photo.jpg --output ./output/
```
## License: MIT
|