Very Deep VAE

Reported on 4 benchmarks across 1 task · 1 paper · 2 SOTA

Note: results are matched by exact model name. Different papers may use the same name for different model variants.

Medical4 results

Image GenerationonFFHQ 256 x 256
bits/dimension· 2020-11-20
0.61
best: 0.53 (Efficient-VDVAE)
SOTA
Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images arXiv:2011.10650
Image GenerationonFFHQ 1024 x 1024
bits/dimension· 2020-11-20
2.42
best: 2.3 (Efficient-VDVAE)
SOTA
Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images arXiv:2011.10650
Image GenerationonImageNet 64x64
Bits per dim· 2020-11-20
3.52
best: 4.351 (Logsparse (6 layers))
Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images arXiv:2011.10650
Image GenerationonImageNet 32x32
bpd· 2020-11-20
3.8
best: 3.34 (NFDM)
Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images arXiv:2011.10650