Novel View Synthesis on DNA-Rendering Diffuman4D 16-scene (test)
25.44PSNRFlex4DHuman-fg
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Flex4DHuman-fgNumber of reference views=1, Number of target views=47, Input reference type=foreground-only (matted reference)2026.06 | 25.44 | 0.9516 | 0.0617 | |
| Flex4DHuman-unmattedNumber of reference views=1, Number of target views=47, Input reference type=whole-scene (full-RGB reference)2026.06 | 25.27 | 0.9268 | 0.0977 | |
| Diffuman4D-GT-skeletonNumber of reference views=1, Number of target views=47, Input type=ground-truth skeleton (oracle)2026.06 | 24.23 | 0.9479 | 0.0744 | |
| MV-PerformerNumber of reference views=1, Number of target views=47, Input type=monocular depth and human normals2026.06 | 17.44 | 0.7204 | 0.2697 | |
| Diffuman4D-mono-skeletonNumber of reference views=1, Number of target views=47, Input type=skeleton lifted from single reference2026.06 | 16.12 | 0.876 | 0.158 |