Image Reconstruction on ImageNet-1k 128x128 (val)
100UtilizationVQGAN-FC (d=8)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VQGAN-FC (d=8)latent dim=8, tokens=256, Codebook size=65,5362026.06 | 100 | 2.63 | 0.13 | 23.79 | 77.5 | |
| FSQlatent dim=6, tokens=256, Codebook size=64,0002026.06 | 100 | 2.8 | 0.13 | 23.63 | 75.8 | |
| LFQlatent dim=16, tokens=256, Codebook size=65,5362026.06 | 100 | 2.88 | 0.13 | 23.6 | 77.2 | |
| VQGAN-LClatent dim=128, tokens=256, Codebook size=65,5362026.06 | 100 | 2.39 | 0.13 | 23.99 | 77.3 | |
| SimVQlatent dim=128, tokens=256, Codebook size=65,5362026.06 | 100 | 2.39 | 0.13 | 24.07 | 77.4 | |
| FVQlatent dim=128, tokens=256, Codebook size=65,5362026.06 | 100 | 2.43 | 0.13 | 24.81 | 77 | |
| NSVQlatent dim=128, tokens=256, Codebook size=65,5362026.06 | 100 | 2.1 | 0.12 | 25.21 | 78.8 | |
| SimVQ + NS loss + freeze-and-refine schedulelatent dim=128, tokens=256, Codebook size=65,5362026.06 | 99.2 | 2.19 | 0.12 | 24.86 | 77.4 | |
| VQGAN-EMAlatent dim=128, tokens=256, Codebook size=65,5362026.06 | 4.5 | 3.23 | 0.15 | 22.89 | 72.3 | |
| VQGANlatent dim=128, tokens=256, Codebook size=65,5362026.06 | 1.4 | 3.74 | 0.17 | 22.2 | 70.6 | |
| VQGAN-FC (d=128)latent dim=128, tokens=256, Codebook size=65,5362026.06 | 1.4 | 5.33 | 0.18 | 21.45 | 68.8 |