3D Human Reconstruction on THuman and Twindom averaged v2.0 (test)
0.122CD (L1)TransferLoss
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TransferLossInput=RGB-D, Views=4, Voxel Resolution=512^32023.04 | 0.122 | 0.917 | 73.6 | — | — | |
| Function4DInput=RGB-D, Views=4, Aggregation=Averaging2023.04 | 0.129 | 0.893 | 70.4 | — | — | |
| MV-PIFuHDInput=RGB-D, Views=4, Resolution=1024x1024, Aggregation=Averaging2023.04 | 0.135 | 0.906 | 69 | — | — | |
| MV-PIFuInput=RGB-D, Views=4, Aggregation=Averaging2023.04 | 0.145 | 0.912 | 62.4 | — | — | |
| FOF-SMPLResolution=512x512x512, Ground-truth SMPL guidance=true2022.06 | 1.58 | — | — | 0.878 | 2.652 | |
| FOF-NormalResolution=512x512x5122022.06 | 2.601 | — | — | 1.435 | 2.837 | |
| FOFResolution=512x512x5122022.06 | 2.693 | — | — | 1.482 | 2.807 | |
| ICONResolution=512x512x512, Ground-truth SMPL guidance=true, Training Status=Pre-trained checkpoint2022.06 | 2.831 | — | — | 1.586 | 3.723 | |
| ICON filterResolution=512x512x512, Ground-truth SMPL guidance=true, Training Status=Pre-trained checkpoint, Note=provided in their code, visually better but may have slightly worse metrics2022.06 | 2.856 | — | — | 1.544 | 3.628 | |
| PIFuHDResolution=512x512x512, Training Status=Pre-trained checkpoint2022.06 | 2.983 | — | — | 1.773 | 2.648 | |
| PIFuResolution=512x512x512, Training Status=Retrained on paper dataset2022.06 | 3.688 | — | — | 2.604 | 3.265 |