Image Generation on ImageNet-1K 512x512 (test)
2.13gFIDTiTok-B-128 + MaskGIT-ViT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TiTok-B-128 + MaskGIT-ViTtokenizer=TiTok-B-128, #tokens=128, codebook size=4096, generator=MaskGIT-ViT [9], P (generator's parameters)=177M, S (sampling steps)=64, T (throughput samples/sec)=7.42024.06 | 2.13 | 1.52 | |
| TiTok-L-64 + MaskGIT-ViT (33.3 FPS)tokenizer=TiTok-L-64, #tokens=64, codebook size=4096, generator=MaskGIT-ViT [9], P (generator's parameters)=177M, S (sampling steps)=8, T (throughput samples/sec)=33.32024.06 | 2.49 | 1.77 | |
| TiTok-L-64 + MaskGIT-ViT (41.0 FPS)tokenizer=TiTok-L-64, #tokens=64, codebook size=4096, generator=MaskGIT-ViT [9], P (generator's parameters)=177M, S (sampling steps)=8, T (throughput samples/sec)=412024.06 | 2.74 | 1.77 | |
| DiT-XL/2tokenizer=VAE [57], #tokens=4096x3, generator=DiT-XL/2 [49], P (generator's parameters)=675M, S (sampling steps)=250, T (throughput samples/sec)=0.12024.06 | 3.04 | 0.19 | |
| UVIT-H/4tokenizer=VAE [57], #tokens=4096x3, generator=UVIT-H/4 [4], P (generator's parameters)=501M, S (sampling steps)=50, T (throughput samples/sec)=0.62024.06 | 4.05 | 0.19 | |
| UVIT-L/4tokenizer=VAE [57], #tokens=4096x3, generator=UVIT-L/4 [4], P (generator's parameters)=287M, S (sampling steps)=50, T (throughput samples/sec)=12024.06 | 4.67 | 0.19 | |
| MaskGIT-ViT with MaskGIT-VQGANtokenizer=MaskGIT-VQGAN [9], #tokens=1024, codebook size=1024, generator=MaskGIT-ViT [9], P (generator's parameters)=177M, S (sampling steps)=12, T (throughput samples/sec)=3.92024.06 | 7.32 | 1.97 |