3D facial animation generation on Held-out set of actors (test)
0.77LVE (mm)KM-Speaker
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| KM-SpeakerGlobal style encoder signal=55 keypoints2026.06 | 0.77 | 0.45 | 0.46 | 4.31 | |
| KM-Speaker (68 keypoints)Global style encoder signal=68 keypoints2026.06 | 0.78 | 0.47 | 0.45 | 4.52 | |
| KM-Speaker (1660 keypoints)Global style encoder signal=1660 keypoints2026.06 | 0.83 | 0.49 | 0.46 | 5.23 | |
| MIMIC_retrainTraining protocol=retrain, Topology=native2026.06 | 0.85 | 0.49 | 0.5 | 4.69 | |
| KM-Speaker (meshtalk encoder)Global style encoder signal=meshtalk encoder2026.06 | 1.1 | 0.65 | 0.5 | 5.63 | |
| MSMD_retrainTraining protocol=retrain2026.06 | 1.11 | 0.65 | 0.56 | 4.51 | |
| MSMD_finetuneTraining protocol=finetune2026.06 | 1.35 | 0.73 | 0.62 | 6.38 | |
| MSMDTraining protocol=original author-released2026.06 | 1.96 | 1.1 | 0.91 | 9.38 | |
| MIMIC_finetuneTraining protocol=finetune, Topology=FLAME2026.06 | 2.41 | 1.97 | 0.47 | 4.59 | |
| MIMICTraining protocol=original author-released2026.06 | 2.48 | 2 | 0.68 | 4.68 |