Image Reconstruction on ImageNet 50k (val)
0.3rFIDVA-VAE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VA-VAE#Images=160M2025.10 | 0.3 | — | — | — | 213.6 | |
| Open-MAGVIT2-I-In1kToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=128 x 128, Codebook Size=2621442024.09 | 0.34 | 27.02 | 100 | — | — | |
| UniTokRes=256, Code Shape=16 × 16, Codebook Size=16,384 × 4†2025.03 | 0.39 | — | — | — | — | |
| VFM-VAE#Images=44M2025.10 | 0.52 | — | — | — | 214.1 | |
| LlamaGenToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=256 x 256, Codebook Size=163842024.09 | 0.59 | 24.45 | — | — | — | |
| SD-VAE#Images=108M2025.10 | 0.62 | — | — | — | 212.1 | |
| SemHiTokRes=384, Code Shape=27 × 27, Codebook Size=196,6082025.03 | 0.66 | — | — | — | — | |
| FQGANRes=256, Code Shape=16 × 16 × 2, Codebook Size=16,384 × 2†2025.03 | 0.94 | — | — | — | — | |
| IBQToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=262144, Codebook Dim=2562024.12 | 1 | — | 0.84 | 0.203 | — | |
| IBQRes=256, Code Shape=16 × 16, Codebook Size=262,1442025.03 | 1 | — | — | — | — | |
| OmiTokenizer-VQToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=256 x 256, Codebook Size=81922024.09 | 1.11 | — | — | — | — | |
| SD-VQGANToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=256 x 256, Codebook Size=163842024.09 | 1.14 | — | — | — | — | |
| SemHiTokRes=256, Code Shape=16 × 16, Codebook Size=196,6082025.03 | 1.16 | — | — | — | — | |
| Open-MAGVIT2-I-In1kToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=2621442024.09 | 1.17 | 22.64 | 100 | — | — | |
| Open-MAGVIT2Token Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=262144, Codebook Dim=02024.12 | 1.17 | — | 1 | 0.2038 | — | |
| VQGANToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=256 x 256, Codebook Size=163842024.09 | 1.19 | 23.38 | — | — | — | |
| VIT-VQGANToken Type=2D, Tokens=32 x 32, Ratio=8, Train Resolution=256 x 256, Codebook Size=81922024.09 | 1.28 | — | — | — | — | |
| IBQToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=2562024.12 | 1.37 | — | 0.96 | 0.2235 | — | |
| IBQRes=256, Code Shape=16 × 16, Codebook Size=16,3842025.03 | 1.37 | — | — | — | — | |
| TokenFlowRes=256, Code Shape=680, Codebook Size=32,7682025.03 | 1.37 | — | — | — | — | |
| Open-MAGVIT2Token Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=02024.12 | 1.58 | — | 1 | 0.2261 | — | |
| Titok-BToken Type=1D, Tokens=64, Train Resolution=256 x 256, Codebook Size=40962024.09 | 1.7 | — | — | — | — | |
| Titok-BToken Type=1D, Tokens=64, Train Resolution=256 x 256, Codebook Size=4096, Codebook Dim=162024.12 | 1.7 | — | — | — | — | |
| Titok-SToken Type=1D, Tokens=128, Train Resolution=256 x 256, Codebook Size=40962024.09 | 1.71 | — | — | — | — | |
| Titok-SToken Type=1D, Tokens=128, Train Resolution=256 x 256, Codebook Size=4096, Codebook Dim=162024.12 | 1.71 | — | — | — | — | |
| VILA-URes=256, Code Shape=16x16x4‡, Codebook Size=16,3842025.03 | 1.8 | — | — | — | — | |
| LlamaGenToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=163842024.09 | 2.19 | 20.79 | 97 | — | — | |
| LlamaGenToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=82024.12 | 2.19 | — | 0.97 | 0.2281 | — | |
| LlamaGenRes=256, Code Shape=16 × 16, Codebook Size=16,3842025.03 | 2.19 | — | — | — | — | |
| TokLIPRes=256, Code Shape=16 × 16, Codebook Size=16,3842025.03 | 2.19 | — | — | — | — | |
| Titok-LToken Type=1D, Tokens=32, Train Resolution=256 x 256, Codebook Size=40962024.09 | 2.21 | — | — | — | — | |
| Titok-LToken Type=1D, Tokens=32, Train Resolution=256 x 256, Codebook Size=4096, Codebook Dim=162024.12 | 2.21 | — | — | — | — | |
| SDE(MUSE-VL)Res=256, Code Shape=16 × 16, Codebook Size=32,7682025.03 | 2.26 | — | — | — | — | |
| MaskGITToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=10242024.09 | 2.28 | — | — | — | — | |
| MaskGITToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=1024, Codebook Dim=2562024.12 | 2.28 | — | — | — | — | |
| VQGAN-LCToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=100000, Codebook Dim=82024.12 | 2.62 | — | 0.99 | 0.2212 | — | |
| VQGAN-LCRes=256, Code Shape=16 × 16, Codebook Size=100,0002025.03 | 2.62 | — | — | — | — | |
| VQGAN-LCToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=82024.12 | 3.01 | — | 0.99 | 0.2358 | — | |
| RQVAERes=256, Code Shape=16 × 16 × 4‡, Codebook Size=16,3842025.03 | 3.2 | — | — | — | — | |
| QLIP-BRes=256, Code Shape=16 × 16, Codebook Size=2^{28}2025.03 | 3.21 | — | — | — | — | |
| VQGAN (reproduced)Token Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=2562024.12 | 3.98 | — | 0.053 | 0.2873 | — | |
| VQGANToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=2562024.12 | 4.98 | — | 0.059 | 0.2843 | — | |
| SD-VQGANToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=163842024.09 | 5.15 | — | — | — | — | |
| SD-VQGANToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=82024.12 | 5.15 | — | — | — | — | |
| VQGANToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=10242024.09 | 7.94 | 19.4 | — | — | — | |
| VQGANToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=1024, Codebook Dim=2562024.12 | 7.94 | — | 0.44 | — | — | |
| LlamaGenToken Type=2D, Tokens=16 x 16, Ratio=16, Train Resolution=256 x 256, Codebook Size=16384, Codebook Dim=2562024.12 | 9.21 | — | 0.0029 | — | — |