Novel View Synthesis on Synthetic Scenes LDR-NE: t2, t4 (test)
43PSNRGaussHDR (Scaffold-GS)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GaussHDR (Scaffold-GS)Training setting=Exp-3, Representation=Scaffold-GS2025.03 | 43 | 0.99 | 0.004 | |
| HDR-Scaffold-GSTraining setting=Exp-3, Representation=Scaffold-GS2025.03 | 42.35 | 0.988 | 0.005 | |
| GaussHDR (3DGS)Training setting=Exp-3, Representation=3DGS2025.03 | 41.57 | 0.985 | 0.007 | |
| HDR-GSTraining setting=Exp-3, Supervision (HDR GT)=false, Implementation=Re-implementation2025.03 | 39.78 | 0.98 | 0.012 | |
| HDR-NeRFTraining setting=Exp-3, Implementation=Re-implementation2025.03 | 39.29 | 0.975 | 0.02 | |
| HDR-NeRF2021.11 | 37.53 | 0.966 | 0.024 | |
| HDR-GSTraining setting=Exp-3, Supervision (HDR GT)=true2025.03 | 36.33 | 0.977 | 0.016 | |
| NeRF-GTTraining data=LDR images with consistent exposures or HDR images, Role=Upper bound2021.11 | 35.87 | 0.955 | 0.032 | |
| NeRF-WNote=Exposures of input views are randomly selected from all five exposures to learn five appearance vectors for testing2021.11 | 29.22 | 0.927 | 0.05 |