Statement · Digital art
From photography to 3D Gaussian splatting
| Aspect | Photogrammetry | LiDAR point clouds | NeRF | 3D Gaussian splatting |
|---|---|---|---|---|
| Input | Multi-view images | Laser scans | Multi-view images | Multi-view images |
| Output | Meshes / textures | Dense point clouds | Volumetric fields | Gaussian splats |
| Accuracy | Moderate (cm-level) | High (mm-level) | Lower | Good |
| Speed | Slow processing | Fast capture | Very slow training | Fast training / render |
| Strengths | Photorealistic textures (PBR materials) | Low-light | Novel views | Real-time novel views |
| Cost / access | Low (cameras) | High (hardware) | Compute-heavy | GPU-efficient |
Generated Memories — Hong Kong
