notesum.ai
Published at December 3Scaling Image Tokenizers with Grouped Spherical Quantization
cs.CV
cs.AI
Released Date: December 3, 2024
Authors: Jiangtao Wang1, Zhen Qin2, Yifan Zhang3, Vincent Tao Hu4, Björn Ommer, Rania Briq, Stefan Kesselheim
Aff.: 1Jülich Supercomputing Centre; 2TapTap; 3Tsinghua University; 4CompVis @ LMU Munich, MCML

| Codebook Init | Norm | rFID | IS | LPIPS | PSNR | SSIM | Usage | PPL |
|---|---|---|---|---|---|---|---|---|
| 11.37 | 84 | 0.12 | 22.3 | 0.64 | 3.38% | 237 | ||
| 5.343 | 113 | 0.10 | 23.7 | 0.71 | 100% | 8077 | ||
| 5.343 | 113 | 0.12 | 23.9 | 0.72 | 100% | 7408 | ||
| 8.312 | 94 | 0.12 | 22.1 | 0.66 | 33.9% | 566 | ||
| 5.375 | 113 | 0.11 | 23.59 | 0.71 | 100% | 8062 |