GS Performance
Evaluation
| scene | machine | image_root | $d^3$ | PSNR↑ | SSIM↑ | LPIPS↓ | PSNR↑ | SSIM↑ | LPIPS↓ | resolution | #gs | times |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Train | Test | |||||||||||
| SAGA | ||||||||||||
| tower_0529 | docker-saga | Drone/Tower_0529$^{gs}$ | 8 | 26.67 | 0.9084 | 0.0800 | 654x490 | 4.97M | ||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{gs}$ | 4 | 23.53 | 0.7965 | 0.1854 | 7.82M | |||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{gs}$ | 2 | 22.13 | 0.7014 | 0.2973 | 2621x1961 | 8.69M | ||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{gs}$ | 1 | 23.43 | 0.7798 | 0.2102 | 1600x1196 | 7.93M | ||||
| GS | ||||||||||||
| tower_0529 | docker-gs | Drone/Tower_0529$^{gs}$ | 8 | 26.80 | 0.9096 | 0.0801 | 654x490 | 3.57M | ||||
| tower_0529 | docker-gs | Drone/Tower_0529$^{gs}$ | 4 | 24.54 | 0.8355 | 0.1568 | 1309x979 | 6.34M | ||||
| 2DGS | ||||||||||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 8 | 26.17 | 2.03M | |||||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 4 | 24.11 | 6.81M | |||||||
| EAGLES | ||||||||||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 8 | 25.36 | 0.8684 | 0.1192 | 1.39M | |||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 4 | 23.59 | 0.7889 | 0.2054 | 2.46M | |||||
| HAC | ||||||||||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 8 | 26.67 | 0.9027 | 0.0882 | 1.20M | |||||
| tower_0529 | lambda-A | Drone/Tower_0529$^{gs}$ | 4 | 24.24 | 0.8191 | 0.1745 | 2.20M | |||||
| NerfStudio-Splatfacto | ||||||||||||
| tower_0529 | docker-nerf-studio | Drone/Tower_0529$^{ns}$ | 8 | 28.95 | 0.9246 | 0.0599 | 624K | |||||
| tower_0529 | docker-nerf-studio | Drone/Tower_0529$^{ns}$ | 4 | 26.84 | 0.8730 | 0.1117 | 1.26M | |||||
| tower_0529 | docker-nerf-studio | Drone/Tower_0529$^{ns}$ | 2 | 24.66 | 0.8016 | 0.1931 | 2.31M | |||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{ns}$ | 1 | 22.52 | 0.7156 | 0.3083 | 3.67M | |||||
| NerfStudio-Splatfacto-Big | ||||||||||||
| tower_0529 | docker-nerf-studio | Drone/Tower_0529$^{ns}$ | 8 | 29.69 | 0.9375 | 0.0485 | 1.72M | |||||
| tower_0529 | docker-nerf-studio | Drone/Tower_0529$^{ns}$ | 4 | 27.57 | 0.8951 | 0.0871 | 3.43M | |||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{ns}$ | 2 | 25.37 | 0.8346 | 0.1469 | 6.30M | |||||
| tower_0529 | lambda-Utha | Drone/Tower_0529$^{ns}$ | 1 | 23.13 | 0.7538 | 0.2423 | 10.31M | |||||
| FreeFloaters | ||||||||||||
| tower_0529 | docker-floater-free | Drone/Tower_0529$^{gs}$ | 8 | 26.97 | 0.9163 | 0.0708 | 4.01M | |||||
| tower_0529 | docker-floater-free | Drone/Tower_0529$^{gs}$ | 4 | 24.96 | 0.8589 | 0.1228 | 7.35M | |||||
| ScaffoldGS | ||||||||||||
| tower_0529 | docker-scaffold-gs | Drone/Tower_0529$^{gs}$ | 8 | 26.58 | 0.9047 | 0.0832 | 702K | |||||
| tower_0529 | docker-scaffold-gs | Drone/Tower_0529$^{gs}$ | 4 | 23.89 | 0.8157 | 0.1683 | 1.06M | |||||
| LightGaussian | ||||||||||||
| tower_0529 | docker-light-gs | Drone/Tower_0529$^{gs}$ | 8 | 25.51 | 0.8754 | 0.1177 | 1.73M | |||||
| EFA-Mip-GS | ||||||||||||
| tower_0529 | 4090 | Drone/Tower_0529$^{gs}$ | 8 | 27.44 | 0.9214 | 0.0663 | 24.27 | 0.8556 | 0.1010 | 2.9M | 21:20 |
- $data^{gs}$ Origianl GS procedure
- $data^{ns}$ NerfStudio procedure
- $d^3$ is the downsampling factor during training.