List your company
ResearchFormerly WizWand

Unconditional Image Generation on FFHQ 256x256 (test)

3.8FIDStyleGAN2
2.8129.48116.1522.819Sep 19, 2022Oct 30, 2022Dec 10, 2022Jan 20, 2023Mar 2, 2023Apr 12, 2023May 23, 2023
Updated 8mo ago

Evaluation results

MethodLinks
StyleGAN2# Params=30M, # Steps=12022.093.8
StyleGAN2Model Type=GAN2023.053.8
LDM#Param=274.1M, MACs=96.1G2022.115
ViT-VQGAN# Params=599M + 1697M, # Steps=10242022.095.3
ViT-VQGANModel Type=Autoregressive, #Params=2.2B2023.055.3
StackformerModel Type=Autoregressive, #Params=651M2023.055.67
Stackformer#Params=(44.4+607)M, Model Size=Base2023.055.67
Stackformer#Params=(44.4+307)M, Model Size=Small2023.056.84
P2#Param=113.7M, MACs=248.7G2022.117
DDPM#Param=113.7M, MACs=248.7G2022.118.4
Mo-VQGAN-auto# Params=82.7M + 307M, # Steps=1024, scenario=auto2022.098.52
Mo-VQGAN#Params=(82.7+307)M2023.058.52
Mo-VQGAN-mask# Params=82.7M + 307M, # Steps=8, scenario=mask2022.098.78
ImageBART2022.099.57
ImageBARTModel Type=Diffusion, #Params=3.5B2023.059.57
RQVAE# Params=100M + 355M, # Steps=2562022.0910.38
RQ-Transformer#Params=(100+355)M2023.0510.38
SD#Param=21.1M, MACs=6.7G2022.1110.5
VQGAN (w/ top-k sampling)# Params=72.1M + 801M, # Steps=2562022.0911.4
VQGAN#Params=(72.1+307)M2023.0511.4
BigGAN# Params=164M, # Steps=12022.0912.4
BigGANModel Type=GAN, #Params=164M2023.0512.4
DCT#Params=738M2023.0513.06
Lite-LDM#Param=22.4M, MACs=7.9G2022.1117.3
VDVAEModel Type=VAE, #Params=115M2023.0528.5