Image Generation on SVHN (test)
13.51FIDIPLD
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| IPLDNumber of particles=10, Number of evaluation samples=50,0002025.05 | 13.51 | — | — | — | — | — | |
| QC-NCSN++NFE=1000, Sampler=PC sampler2022.09 | 13.88 | — | 3.12 | 61 | 67 | — | |
| IPLDNumber of particles=5, Number of evaluation samples=50,0002025.05 | 14.02 | — | — | — | — | — | |
| U-NCSN++NFE=1000, Sampler=PC sampler2022.09 | 14.34 | — | 3.1 | 60 | 67 | — | |
| IPLDNumber of particles=1, Number of evaluation samples=50,0002025.05 | 17.55 | — | — | — | — | — | |
| Ours-DAMC2023.10 | 18.76 | — | — | — | — | 0.002 | |
| DAMCNumber of evaluation samples=50,0002025.05 | 18.76 | — | — | — | — | — | |
| DiffusionVAE*Number of evaluation samples=50,000, re-implementation=true2025.05 | 20.89 | — | — | — | — | — | |
| Ours-LEBM2023.10 | 21.17 | — | — | — | — | 0.002 | |
| C-NCSN++NFE=1000, Sampler=PC sampler2022.09 | 24.71 | — | 2.66 | 61 | 46 | — | |
| Adaptive CEPrior model parameters=4x compared with baseline2023.10 | 26.19 | — | — | — | — | 0.004 | |
| LEBM2023.10 | 29.44 | — | — | — | — | 0.008 | |
| LP-EBMNumber of evaluation samples=50,0002025.05 | 29.44 | — | — | — | — | — | |
| NCP-VAE2023.10 | 33.23 | — | — | — | — | 0.02 | |
| SRIL=52023.10 | 35.32 | — | — | — | — | 0.011 | |
| RAE2023.10 | 40.02 | — | — | — | — | 0.014 | |
| 2s-VAE2023.10 | 42.81 | — | — | — | — | 0.019 | |
| SRI2023.10 | 44.86 | — | — | — | — | 0.018 | |
| VAE2023.10 | 46.78 | — | — | — | — | 0.019 | |
| ABP2023.10 | 49.71 | — | — | — | — | — | |
| HoffmanModel Architecture=Fully connected deep neural networks2022.09 | — | 4.44 | — | — | — | — | |
| LAEALD iterations (T)=2, sigma=0.05, Model Architecture=Fully connected deep neural networks2022.09 | — | 4.412 | — | — | — | — | |
| VAEModel Architecture=Fully connected deep neural networks2022.09 | — | 4.442 | — | — | — | — | |
| VAE-flowModel Architecture=Fully connected deep neural networks2022.09 | — | 4.454 | — | — | — | — |