Image Reconstruction on FFHQ 10,000 images (val)
2.26rFIDMo-VQGAN
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Mo-VQGANLatent Size=16x16x4, Num Z=10242022.09 | 2.26 | 26.72 | 0.8212 | 0.0585 | |
| ViT-VQGANLatent Size=32x32, Num Z=81922022.09 | 3.13 | — | — | — | |
| RQ-VAELatent Size=16x16x4, Num Z=2048, trained with publicly available code=true2022.09 | 3.88 | 24.53 | 0.7602 | 0.0895 | |
| VQGANLatent Size=16x16, Num Z=10242022.09 | 4.42 | 22.24 | 0.6641 | 0.1175 | |
| RQ-VAELatent Size=8x8x4, Num Z=20482022.09 | 7.04 | 22.99 | 0.67 | 0.1302 |