List your company

Class conditional image generation on ImageNet

2.21FIDSODA
1.9843.50955.0356.5605Oct 26, 2019Dec 4, 2020Jan 14, 2022Feb 24, 2023Apr 4, 2024May 15, 2025Jun 25, 2026

Evaluation results

MethodLinks
SODASampling=DDPM, Ns=1252026.032.21272.01—8258—4.72————1.77876.71.55
DiT-XL/2Sampling=DDPM, Steps=2502026.032.23275.65—8258—4.57————2.508118.681
DiT-XL/2Sampling=DDIM, Steps=502026.032.25239.97—8059—4.33————0.53323.741
DDIM-50 stepsBackbone=DiT-XL/2, Sampling steps=50, MACs (T)=11.44, Speedup=1.00×2026.042.26——8060—4.29———————
Full-ComputeNFC=50, FLOPs (T)=23.74, Latency (s)=7.612026.022.32241.25————4.32———————
ResilPhaseN=3, O=4, Speed=2.78×, Backbone=DiT-XL/22026.062.347233.57————4.672———————
DDIMSampling steps=50, Speed=1.00×, Backbone=DiT-XL/22026.062.367236.06————4.387———————
SODASampling=DDIM, Ns=232026.032.39240.75—8158—4.55————0.31412.211.94
DDIM-40 stepsBackbone=DiT-XL/2, Sampling steps=40, MACs (T)=9.14, Speedup=1.24×2026.042.39——8059—4.28———————
SODASampling=DDPM, Ns=722026.032.47262.3—8159—5.09————1.0743.422.73
RFCm=2, NFC=14.01, FLOPs (T)=6.67, Latency (s)=2.862026.022.52231————4.6——0.31.33———
TaylorSeerN=4, m=2, NFC=14, FLOPs (T)=6.66, Latency (s)=2.842026.022.55232.26————5.3——0.442.17———
ToCaSampling=DDPM2026.032.58256.26—8059—5.74————1.21143.222.75
DuCaSampling=DDPM2026.032.59256.36—8059—5.68————1.1743.422.73
SpeCaτ0=0.1, β=0.5, Speed=2.65×, Backbone=DiT-XL/22026.062.633231.32————4.848———————
TaylorSeerN=5, m=2, NFC=11, FLOPs (T)=5.24, Latency (s)=2.522026.022.7228.76————5.45——0.562.74———
RFCm=2, NFC=11.01, FLOPs (T)=5.24, Latency (s)=2.522026.022.71227.72————4.8——0.482.14———
RFCm=2, NFC=10.04, FLOPs (T)=4.78, Latency (s)=2.462026.022.75226.93————5.02——0.542.43———
SODASampling=DDIM, Ns=182026.032.75235.65—8058—4.56————0.2639.552.49
FORASampling=DDPM2026.032.8——8059—6.13—————43.362.74
DiffSparseBackbone=DiT-XL/2, Sampling steps=50, MACs (T)=4.97, Speedup=2.07×2026.042.81——8059—4.61———————
ResilPhaseN=5, O=3, Speed=4.41×, Backbone=DiT-XL/22026.062.832219.19————5.026———————
SODASampling=DDIM, Ns=162026.032.88234.91—8058—4.71————0.2188.432.82
DDIM-25 stepsBackbone=DiT-XL/2, Sampling steps=25, MACs (T)=5.73, Speedup=1.96×2026.043.01——7958—4.6———————
ToCaN=4, NFC=17, FLOPs (T)=8.73, Latency (s)=3.642026.023.03229.01————5.02——0.863.56———
DuCaBackbone=DiT-XL/2, Sampling steps=50, MACs (T)=4.94, Speedup=2.10×2026.043.04——7957—4.7———————
ToCaSampling=DDIM2026.033.05——7957—4.7————0.28910.232.32
DuCaSampling=DDIM2026.033.05233.21—8058—4.66————0.2769.582.48
ToCaBackbone=DiT-XL/2, Sampling steps=50, MACs (T)=4.97, Speedup=2.09×2026.043.05——7957—4.7———————
TaylorSeerN=6, m=2, NFC=10, FLOPs (T)=4.76, Latency (s)=2.272026.023.1222.85————6.47——14.63———
RFCm=2, NFC=8.02, FLOPs (T)=3.83, Latency (s)=2.152026.023.12219.2————5.07——0.813.1———
ClusCaN=6, K=8, O=4, Speed=2.61×, Backbone=DiT-XL/22026.063.251217.6————5.207———————
DuCaN=4, NFC=17, FLOPs (T)=7.66, Latency (s)=2.942026.023.39224.03————4.95——1.194.22———
RFCm=2, NFC=7.04, FLOPs (T)=3.37, Latency (s)=1.992026.023.4215.39————5.21——1.033.66———
TaylorSeerN=7, m=2, NFC=8, FLOPs (T)=3.82, Latency (s)=2.022026.023.46217.18————6.97——1.35.61———
DDIM-20 stepsBackbone=DiT-XL/2, Sampling steps=20, MACs (T)=4.58, Speedup=2.42×2026.043.48——7956—4.64———————
ToCaN=5, NFC=14, FLOPs (T)=7.43, Latency (s)=3.352026.023.53224.07————5.36——1.284.5———
ToCaN=6, NFC=13, FLOPs (T)=7.01, Latency (s)=3.272026.023.54226.32————5.54——1.445.18———
DiT-XL/2Sampling Steps=20, Backbone=DiT-XL/22026.033.56221.27—7858——————1.719.49—
LDM-4-GNparams=400M, Sampling steps=250 steps, Classifier-free guidance (c.f.g.)=true, Guidance scale (s)=1.52021.123.6——8748—————————
DDIMSampling steps=20, Speed=2.50×, Backbone=DiT-XL/22026.063.858216.76————5.201———————
FORABackbone=DiT-XL/2, Sampling steps=50, MACs (T)=4.13, Speedup=2.12×2026.043.88——7956—6.74———————
ToCaN=7, NFC=14, FLOPs (T)=6.25, Latency (s)=2.92026.024.04214.25————5.72——1.624.72———
GENIEAFS=false, Denoising=true, NFEs=252022.104.13—————————————
ViT-VQGANModel Type=Autoregressive, #Params=2.2B2023.054.17175.1————————————
GENIEAFS=true, Denoising=true, NFEs=252022.104.27—————————————
FORAN=4, NFC=14, FLOPs (T)=6.66, Latency (s)=2.372026.024.33215.33————7.38——1.96.32———
GENIEAFS=false, Denoising=true, NFEs=202022.104.48—————————————
ADM-Gclassifier guidance=1.02021.114.59—————————————
ADM-GNparams=608M, Sampling steps=250 DDIM steps2021.124.59186.7—8252—————————
GENIEAFS=true, Denoising=true, NFEs=202022.104.68—————————————
FORAN=4, Speed=2.60×, Backbone=DiT-XL/22026.064.77210.98————7.591———————
TaylorSeerN=8, O=3, Speed=2.14×, Backbone=DiT-XL/22026.064.807197.22————7.088———————
TaylorSeerN=9, m=2, NFC=7, FLOPs (T)=3.35, Latency (s)=1.892026.024.9198.46————7.92——2.337.35———
GENIEAFS=false, Denoising=false, NFEs=252022.105—————————————
DDIMAFS=false, Denoising=true, NFEs=252022.105.19—————————————
GENIEAFS=false, Denoising=true, NFEs=152022.105.23—————————————
GENIEAFS=true, Denoising=false, NFEs=252022.105.27—————————————
VQ-Diffusionacceptance rate=0.052021.115.32—————————————
DDIMAFS=true, Denoising=true, NFEs=252022.105.35—————————————
GENIEAFS=false, Denoising=false, NFEs=202022.105.36—————————————
GENIEAFS=true, Denoising=true, NFEs=152022.105.36—————————————
GENIEAFS=true, Denoising=false, NFEs=202022.105.73—————————————
DDIMAFS=false, Denoising=true, NFEs=202022.105.83—————————————
VQGANacceptance rate=0.052021.115.88—————————————
DDIMAFS=true, Denoising=true, NFEs=202022.105.93—————————————
DuCaN=5, NFC=14, FLOPs (T)=6.32, Latency (s)=2.572026.026.02196.96————6.71——4.198.18———
StackformerModel Type=Autoregressive, #Params=651M2023.056.04172.6————————————
FORAN=5, NFC=11, FLOPs (T)=5.24, Latency (s)=2.032026.026.12196.04————10.51——3.5710.19———
GENIEAFS=false, Denoising=false, NFEs=152022.106.13—————————————
ToCaN=5, R=93%, Speed=2.67×, Backbone=DiT-XL/22026.066.321196.06————7.027———————
DuCaN=6, NFC=13, FLOPs (T)=5.86, Latency (s)=2.532026.026.37187.74————6.74——4.517.86———
GENIEAFS=true, Denoising=false, NFEs=152022.106.59—————————————
DDIMAFS=false, Denoising=false, NFEs=252022.106.64—————————————
CR-BigGAN*Noise truncation=false, Consistency regularization=true2019.106.66—————————————
BigGAN-deep2021.116.84—————————————
SpeCaτ0=1.5, β=0.5, Speed=4.36×, Backbone=DiT-XL/22026.066.866171.68————8.25———————
DDIMAFS=true, Denoising=false, NFEs=252022.106.92—————————————
TC-PadéCache Type=Prediction-based, Mode=fast, Backbone=DiT-XL/22026.036.93185.12—7254——————1.175.79—
S-PNDMAFS=false, Denoising=true, NFEs=252022.106.94—————————————
S-PNDMAFS=true, Denoising=true, NFEs=252022.106.94—————————————
DDIM-GMM-ORTHO-VUBSteps=10, Guidance Scale=2.52023.116.94207.85————————————
BigGan-deepNparams=340M2021.126.95——8728—————————
DDIMAFS=true, Denoising=true, NFEs=152022.107.14—————————————
DDIMAFS=false, Denoising=true, NFEs=152022.107.17—————————————
S-PNDMAFS=false, Denoising=false, NFEs=252022.107.2—————————————
S-PNDMAFS=true, Denoising=false, NFEs=252022.107.22—————————————
GENIEAFS=true, Denoising=true, NFEs=102022.107.41—————————————
ImageBARTacceptance rate=0.052021.117.44—————————————
S-PNDMAFS=true, Denoising=true, NFEs=202022.107.52—————————————
BigGAN2021.117.53—————————————
RQ-TransformerModel Type=Autoregressive, #Params=3.8B2023.057.55134————————————
DDIMAFS=false, Denoising=false, NFEs=202022.107.57—————————————
GENIEAFS=false, Denoising=true, NFEs=102022.107.59—————————————
S-PNDMAFS=false, Denoising=true, NFEs=202022.107.59—————————————
F-PNDMAFS=true, Denoising=false, NFEs=252022.107.74—————————————
BigGAN*Noise truncation=false, Implementation=Kurach et al. (2019)2019.107.75—————————————
DuCaN=7, NFC=14, FLOPs (T)=4.96, Latency (s)=2.242026.027.76181.5————7.77——5.458.8———
TaylorSeerCache Type=Prediction-based, N (Cache interval)=3, O (Order of Taylor expansion)=2, Backbone=DiT-XL/22026.037.84175.99—7153——————1.193.8—
DDIMAFS=true, Denoising=false, NFEs=202022.107.86—————————————