List your company

Class-conditional Image Generation on ImageNet 64x64 (test)

1.01FIDEDM2*
0.39844.52678.65512.7833May 31, 2024Sep 21, 2024Jan 13, 2025May 7, 2025Aug 29, 2025Dec 21, 2025Apr 14, 2026

Evaluation results

MethodLinks
EDM2*NFE=63, Auto-Guidance=true2025.091.01———
GDD-INFE=12024.051.167560—
RINNFE=10002024.051.23———
RINNFE=10002025.091.23———
AYFNFE=22025.091.25———
DMD2NFE=12025.091.28———
CMT (w/ ECD)NFE=22025.091.32———
EDM2-XLNFE=632024.061.33———
EDM2NFE=632025.091.33———
GDDNFE=12024.051.427759—
sCTNFE=22025.091.48———
CMT (w/ ECT)NFE=22025.091.48———
ECDNFE=22025.091.5———
StyleGAN2-XLNFE=12024.051.51———
SiDNFE=1, alpha=1.22024.051.527463—
SiDNFE=12025.091.52———
EDMType=Diff/FM2024.121.55———
Stable CTNFE=22025.091.55———
TCMNFE=22025.091.62———
sCDNFE=22025.091.66———
ECM-XL*Training budget=102.4M images, NFE=22024.061.67———
ECTNFE=22025.091.67———
CTMNFE=22024.051.737457—
CTMNFE=22025.091.73———
CMT (w/ ECD)NFE=12025.091.78———
MultiStep-CDNFE=22025.091.9———
CTMNFE=12024.051.927357—
CTMNFE=12025.091.92———
ECM-XLTraining iterations=100k, NFE=22024.061.96———
SiDNFE=1, alpha=12024.052.027363—
CMT (w/ ECT)NFE=12025.092.02———
sCTNFE=12025.092.04———
ADMNFE=2502024.052.077463—
ADMNFE=2502024.062.07———
ADM(dropout)Type=Diff/FM2024.122.09———
ECM-LTraining iterations=100k, NFE=22024.062.14———
TCMNFE=12025.092.2———
EDMNFE=792024.062.22———
ECDNFE=12025.092.24———
ECM-MTraining iterations=100k, NFE=22024.062.35———
Stable CTNFE=12025.092.42———
sCDNFE=12025.092.44———
ECM-XL*Training budget=102.4M images, NFE=12024.062.49———
ECTNFE=12025.092.49———
Separate CompressionRetention Ratio=1.0x, MACs=107G, Training Overhead=400N2026.042.57———
DMDNFE=12024.052.62———
EDMNFE=792024.052.64———
TARFLOWType=NF, Patch Size (P)=2, Channel Size (Ch)=1024, Flow Blocks (T)=8, Attention Layers (K)=8, Input Noise (pe)=N(0, 0.05^2)2024.122.66———
iCT-deepNFE=22024.052.777462—
iCT-deepNFE=22024.062.77———
iCT-deepNFE=22025.092.77———
ECM-S*Training budget=102.4M images, NFE=22024.062.79———
TARFLOWType=NF, Patch Size (P)=2, Channel Size (Ch)=768, Flow Blocks (T)=8, Attention Layers (K)=8, Input Noise (pe)=N(0, 0.05^2)2024.122.9———
iDDPMType=Diff/FM2024.122.92———
AYFNFE=12025.092.98———
VCTNFE=22025.093.07———
ECM-STraining iterations=100k, NFE=22024.063.18———
iCTNFE=22024.053.27363—
iCTNFE=22024.063.2———
iCTNFE=22025.093.2———
MultiStep-CDNFE=12025.093.2———
iCT-deepNFE=12024.053.257263—
iCT-deepType=CM2024.123.25———
iCT-deepNFE=12025.093.25———
OFA-Diffusion CompressionFine-tune=40K, Retention Ratio=1.0x, MACs=107G, Training Overhead=400 + 40N2026.043.65———
TARFLOWType=NF, Patch Size (P)=4, Channel Size (Ch)=1024, Flow Blocks (T)=8, Attention Layers (K)=8, Input Noise (pe)=N(0, 0.05^2)2024.123.99———
iCTNFE=12024.054.027063—
iCTNFE=12025.094.02———
ECM-S*Training budget=102.4M images, NFE=12024.064.05———
BigGAN-deepNFE=12024.054.067948—
BigGANType=GAN2024.124.06———
CDNFE=22024.054.76964—
CD (LPIPS)NFE=22024.064.7———
CD(LPIPS)Type=CM2024.124.7———
CDNFE=22025.094.7———
VCTNFE=12025.094.93———
TRACTNFE=22024.064.97———
Diff-InstructNFE=12024.065.57———
PD (LPIPS)NFE=22024.065.74———
CDNFE=12024.056.26863—
CDNFE=12025.096.2———
IC-GANType=GAN2024.126.7———
DFNONFE=12024.057.83—61—
DFNO (LPIPS)NFE=12024.067.83———
OFA-Diffusion CompressionFine-tune=40K, Retention Ratio=0.25x, MACs=32.35G, Training Overhead=400 + 40N2026.048.65———
Separate CompressionRetention Ratio=0.25x, MACs=32.35G, Training Overhead=400N2026.049.45———
DDPMNFE=2502024.05116758—
OFA-Diffusion CompressionRetention Ratio=0.25x, MACs=32.35G, Training Overhead=4002026.0411.35———
PDNFE=12024.0515.39———
BOOTNFE=12024.0516.36836—
BOOTNFE=12024.0616.3———
Naive OFATraining Overhead=4002026.04———8.36
OFA-Diffusion CompressionTraining Overhead=4002026.04———7.43
OFA-Diffusion CompressionFine-tune=40K, Training Overhead=400 + 40N2026.04———5.47
Separate CompressionTraining Overhead=400N2026.04———6.08