Identity-Preserving Text-to-Video generation on IPT2V (evaluation)
0.73FaceSim (Arc)ConsisID
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ConsisIDFull configuration=true2024.11 | 0.73 | 0.75 | 36.77 | 127.42 | |
| w/o LFELocal Facial Extractor=Disabled2024.11 | 0.66 | 0.68 | 34.48 | 104.34 | |
| w/o DCLDynamic cross-face loss=Disabled2024.11 | 0.65 | 0.69 | 32.21 | 117.8 | |
| w/o DMLDynamic mask loss=Disabled2024.11 | 0.62 | 0.67 | 34.23 | 187.78 | |
| w/o CFTCoarse-to-fine training=Disabled2024.11 | 0.54 | 0.58 | 34.47 | 144.62 | |
| w/o GFEGlobal Facial Extractor=Disabled2024.11 | 0.05 | 0.05 | 34.86 | 269.88 |