Image Reconstruction on ImageNet 256x256 2012 (test val)
0.75rFIDCaTok-L-256
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CaTok-L-256Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=16, #Param.=552M, Epochs=160, VQ=×2026.03 | 0.75 | 22.53 | 67.4 | |
| Semanticist-L-256Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=16, #Param.=552M, Epochs=400, VQ=×2026.03 | 0.78 | 21.61 | 62.6 | |
| FlowMo-Lo-256Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=-, #Param.=945M, Epochs=130, VQ=✓2026.03 | 0.95 | 22.07 | 64.9 | |
| One-D-Piece-B-256Tokenizer Type=One-step 1D tokenizer, Token=256, #Dim.=16, #Param.=172M, Epochs=300, VQ=✓2026.03 | 1.11 | 18.77 | — | |
| OpenMagViT-V2Tokenizer Type=One-step 2D tokenizer, Token=16x16, #Dim.=-, #Param.=116M, Epochs=270, VQ=✓2026.03 | 1.17 | 21.63 | 64 | |
| CaTok-B-256Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=16, #Param.=224M, Epochs=80, VQ=×2026.03 | 1.17 | 22.1 | 66.6 | |
| CaTok-L-128Tokenizer Type=Diffusion tokenizer, Token=128, #Dim.=16, #Param.=552M, Epochs=160, VQ=×2026.03 | 1.17 | 21.01 | 60.9 | |
| MAR-VAETokenizer Type=One-step 2D tokenizer, Token=16x16, #Dim.=16, #Param.=-, Epochs=-, VQ=×2026.03 | 1.22 | — | — | |
| Semanticist-L-128Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=16, #Param.=552M, Epochs=400, VQ=×, Trained by official codebase=true2026.03 | 1.24 | 19.59 | 58.6 | |
| MaskBitTokenizer Type=One-step 2D tokenizer, Token=16x16, #Dim.=12, #Param.=54M, Epochs=270, VQ=✓2026.03 | 1.37 | 21.5 | 56 | |
| FlexTok d18-d28Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=6, #Param.=1.4B, Epochs=640, VQ=✓2026.03 | 1.45 | 18.53 | 46.5 | |
| FlexTok d18-d18Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=6, #Param.=573M, Epochs=640, VQ=✓2026.03 | 1.61 | — | — | |
| CaTok-L-64Tokenizer Type=Diffusion tokenizer, Token=64, #Dim.=16, #Param.=552M, Epochs=160, VQ=×2026.03 | 1.61 | 19.36 | 53.3 | |
| TiTok-S-128Tokenizer Type=One-step 1D tokenizer, Token=128, #Dim.=16, #Param.=44M, Epochs=300, VQ=✓2026.03 | 1.71 | 17.52 | 43.7 | |
| CaTok-L-32Tokenizer Type=Diffusion tokenizer, Token=32, #Dim.=16, #Param.=552M, Epochs=160, VQ=×2026.03 | 2.03 | 17.85 | 46.5 | |
| LlamaGenTokenizer Type=One-step 2D tokenizer, Token=16x16, #Dim.=8, #Param.=72M, Epochs=40, VQ=✓2026.03 | 2.19 | 20.67 | 58.9 | |
| TiTok-L-32Tokenizer Type=One-step 1D tokenizer, Token=32, #Dim.=8, #Param.=614M, Epochs=300, VQ=✓2026.03 | 2.21 | 15.6 | 35.9 | |
| SpectralAR-64Tokenizer Type=One-step 1D tokenizer, Token=64, #Dim.=16, #Param.=172M, Epochs=300, VQ=✓2026.03 | 4.03 | — | — | |
| FlexTok d12-d12Tokenizer Type=Diffusion tokenizer, Token=256, #Dim.=6, #Param.=170M, Epochs=640, VQ=✓2026.03 | 4.2 | — | — | |
| CaTok-L-32Tokenizer Type=One-step 1D tokenizer, Token=32, #Dim.=16, #Param.=552M, Epochs=160, VQ=×, One-step sampling=true2026.03 | 4.48 | 17.25 | 44.1 | |
| CaTok-L-256Tokenizer Type=One-step 1D tokenizer, Token=256, #Dim.=16, #Param.=552M, Epochs=160, VQ=×, One-step sampling=true2026.03 | 4.63 | 20.99 | 63 | |
| CaTok-B-256Tokenizer Type=One-step 1D tokenizer, Token=256, #Dim.=16, #Param.=224M, Epochs=80, VQ=×, One-step sampling=true2026.03 | 4.89 | 20.77 | 61.7 | |
| CaTok-L-128Tokenizer Type=One-step 1D tokenizer, Token=128, #Dim.=16, #Param.=552M, Epochs=160, VQ=×, One-step sampling=true2026.03 | 4.92 | 20.36 | 59 | |
| CaTok-L-64Tokenizer Type=One-step 1D tokenizer, Token=64, #Dim.=16, #Param.=552M, Epochs=160, VQ=×, One-step sampling=true2026.03 | 4.96 | 18.56 | 50.6 | |
| VQGANTokenizer Type=One-step 2D tokenizer, Token=16x16, #Dim.=16, #Param.=307M, Epochs=-, VQ=✓2026.03 | 7.94 | — | — | |
| SelfTok-512Tokenizer Type=Diffusion tokenizer, Token=512, #Dim.=16, #Param.=-, Epochs=-, VQ=✓2026.03 | — | 21.86 | 60 |