Image Reconstruction on ImageNet 512x512 (test val)
0.19rFIDSSDD-L
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SSDD-LDecoder Parameters (#D)=85.2M, Encoder=f8c4, Distillation strategy=single-step distilled2025.10 | 0.19 | 0.055 | |
| SSDD-MDecoder Parameters (#D)=48.0M, Encoder=f8c4, Distillation strategy=single-step distilled2025.10 | 0.2 | 0.057 | |
| SSDD-XLDecoder Parameters (#D)=153.8M, Encoder=f8c4, Distillation strategy=single-step distilled2025.10 | 0.2 | 0.053 | |
| SSDD-BDecoder Parameters (#D)=20.2M, Encoder=f8c4, Distillation strategy=single-step distilled2025.10 | 0.22 | 0.06 | |
| DiTo-XL + LPIPSDecoder Parameters (#D)=592.2M, Encoder=f8c42025.10 | 0.28 | 0.08 | |
| SSDD-SDecoder Parameters (#D)=13.4M, Encoder=f8c4, Distillation strategy=single-step distilled2025.10 | 0.28 | 0.062 | |
| KL-VAEDecoder Parameters (#D)=47.2M, Encoder=f8c42025.10 | 0.29 | 0.064 | |
| Consistency decoderDecoder Parameters (#D)=625.1M, Encoder=f8c42025.10 | 0.3 | 0.066 | |
| ε-VAE (H)Decoder Parameters (#D)=355.6M, Encoder=f8c42025.10 | 0.35 | — | |
| ε-VAE (XL)Decoder Parameters (#D)=140.6M, Encoder=f8c42025.10 | 0.39 | — | |
| ε-VAE (L)Decoder Parameters (#D)=89.0M, Encoder=f8c42025.10 | 0.41 | — | |
| ε-VAE (M)Decoder Parameters (#D)=49.3M, Encoder=f8c42025.10 | 0.53 | — | |
| ε-VAE (B)Decoder Parameters (#D)=20.6M, Encoder=f8c42025.10 | 0.61 | — |