Blender
Blender (nerf-synthetic) is a synthetic dataset used to benchmark NeRF methods. It consists of 8 scenes of an object placed on a white background. Cameras are placed on a semi-sphere around the object.
Authors: Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, Ren Ng
Method | PSNR | SSIM | LPIPS | Time | GPU Mem. | |
|---|---|---|---|---|---|---|
| Zip-NeRF | 5h 21m 57s | 26.20 GB | ||||
| Mip-Splatting | 6m 49s | 2.65 GB | ||||
| Gaussian Splatting | 6m 6s | 3.08 GB | ||||
| Instant NGP | 2m 23s | 2.57 GB | ||||
| Tetra-NeRF | 6h 53m 20s | 29.57 GB | ||||
| TensoRF | 8m 9s | 16.86 GB | ||||
| Mip-NeRF 360 | 3h 29m 39s | 114.80 GB | ||||
| NerfStudio | 9m 38s | 3.65 GB |
PSNR
Peak Signal to Noise Ratio. The higher the better.
Method | lego | drums | ficus | hotdog | materials | mic | ship | chair |
|---|---|---|---|---|---|---|---|---|
| Zip-NeRF | ||||||||
| Mip-Splatting | ||||||||
| Gaussian Splatting | ||||||||
| Instant NGP | ||||||||
| Tetra-NeRF | ||||||||
| TensoRF | ||||||||
| Mip-NeRF 360 | ||||||||
| NerfStudio |
SSIM
Structural Similarity Index. The higher the better. The implementation matches JAX's SSIM and torchmetrics's SSIM (with default parameters).
Method | lego | drums | ficus | hotdog | materials | mic | ship | chair |
|---|---|---|---|---|---|---|---|---|
| Zip-NeRF | ||||||||
| Gaussian Splatting | ||||||||
| Mip-Splatting | ||||||||
| Instant NGP | ||||||||
| Tetra-NeRF | ||||||||
| TensoRF | ||||||||
| Mip-NeRF 360 | ||||||||
| NerfStudio |
LPIPS
Learned Perceptual Image Patch Similarity. The lower the better. The implementation uses VGG backbone and matches lpips pip package with checkpoint version 0.1
Method | lego | drums | ficus | hotdog | materials | mic | ship | chair |
|---|---|---|---|---|---|---|---|---|
| Zip-NeRF | ||||||||
| Gaussian Splatting | ||||||||
| Mip-Splatting | ||||||||
| Tetra-NeRF | ||||||||
| Instant NGP | ||||||||
| TensoRF | ||||||||
| Mip-NeRF 360 | ||||||||
| NerfStudio |