Visual Reconstruction on ImageNet-1K (val)
0.88rFIDVanilla CVQ
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Vanilla CVQToken Budget=1024 tokens, # Tokens=1024, Ratio=16, Codebook Size=16,3842026.05 | 0.88 | 0.723 | 25.02 | 100 | |
| MoVQToken Budget=1024 tokens, # Tokens=16×16×4, Ratio=16, Codebook Size=16,3842026.05 | 1.05 | — | — | — | |
| VQ w/ rotation trickToken Budget=1024 tokens, # Tokens=32×32, Ratio=8, Codebook Size=16,3842026.05 | 1.1 | — | — | 27 | |
| MoVQToken Budget=1024 tokens, # Tokens=16×16×4, Ratio=16, Codebook Size=1,0242026.05 | 1.12 | 0.673 | 22.42 | — | |
| VQ-FCToken Budget=1024 tokens, # Tokens=32×32, Ratio=8, Codebook Size=8,1922026.05 | 1.28 | — | — | 95 | |
| VQ-LCToken Budget=1024 tokens, # Tokens=32×32, Ratio=8, Codebook Size=100,0002026.05 | 1.29 | 0.716 | — | 99.5 | |
| Vanilla VQToken Budget=1024 tokens, # Tokens=32×32, Ratio=8, Codebook Size=16,384, Reproduced=true2026.05 | 1.32 | 0.692 | 23.35 | 2.8 | |
| Vanilla VQToken Budget=1024 tokens, # Tokens=32×32, Ratio=8, Codebook Size=8,1922026.05 | 1.49 | — | — | — | |
| RVQToken Budget=1024 tokens, # Tokens=8×8×16, Ratio=32, Codebook Size=16,3842026.05 | 1.83 | — | — | — | |
| Vanilla CVQToken Budget=256 tokens, # Tokens=256, Ratio=16, Codebook Size=16,3842026.05 | 2.6 | 0.565 | 20.94 | 100 | |
| VQ-LCToken Budget=256 tokens, # Tokens=16×16, Ratio=16, Codebook Size=100,0002026.05 | 2.62 | 0.589 | — | 99 | |
| SimVQToken Budget=256 tokens, # Tokens=16×16, Ratio=16, Codebook Size=16,384, Reproduced=true2026.05 | 2.63 | 0.556 | 21.1 | 100 | |
| VQ-LCToken Budget=256 tokens, # Tokens=16×16, Ratio=16, Codebook Size=16,3842026.05 | 3.01 | 0.564 | — | 99 | |
| Dynamic VQToken Budget=256 tokens, # Tokens=16×16, Ratio={8, 16, 32}, Codebook Size=16,384, Reproduced=true2026.05 | 3.62 | 0.558 | 20.07 | 5.2 | |
| Dynamic VQToken Budget=256 tokens, # Tokens=16×16, Ratio={8, 16, 32}, Codebook Size=1,0242026.05 | 4.08 | — | — | — | |
| Vanilla VQToken Budget=256 tokens, # Tokens=16×16, Ratio=16, Codebook Size=16,384, Reproduced=true2026.05 | 4.84 | 0.542 | 19.93 | 4.5 |