Face Reconstruction on Synthetic Face Dataset
9.006AccuracyVGGT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VGGT2026.04 | 9.006 | 4.965 | 80.74 | |
| PiSSA# Trainable Param.=130.7 M, rank=1282026.04 | 6.89 | 4.706 | 84.77 | |
| LoRA# Trainable Param.=130.7 M, rank=1282026.04 | 6.59 | 4.242 | 84.92 | |
| PiSSA# Trainable Param.=65.3 M, rank=642026.04 | 6.526 | 4.407 | 84.84 | |
| PiSSA# Trainable Param.=32.7 M, rank=322026.04 | 6.488 | 4.198 | 84.64 | |
| LoRA# Trainable Param.=65.3 M, rank=642026.04 | 6.393 | 3.971 | 84.59 | |
| LoRA# Trainable Param.=32.7 M, rank=322026.04 | 6.251 | 3.841 | 84.64 | |
| Subspace-based LoRA# Trainable Param.=3.8 M, d=82026.04 | 5.921 | 1.966 | 76.77 | |
| LoRA# Trainable Param.=16.3 M, rank=162026.04 | 5.767 | 3.385 | 84.78 | |
| PiSSA# Trainable Param.=16.3 M, rank=162026.04 | 5.729 | 3.532 | 85.3 | |
| Full# Trainable Param.=853.6 M2026.04 | 5.585 | 3.531 | 85.77 | |
| Subspace-based LoRA# Trainable Param.=4.7 M, d=322026.04 | 4.287 | 2.395 | 86.43 | |
| Subspace-based LoRA# Trainable Param.=4.0 M, d=162026.04 | 3.831 | 2.037 | 86.65 |