List your company

Class-conditional image generation on ImageNet-1K 256x256 (test)

1.5FIDRAR-XL
1.0484.0997.1510.201Nov 27, 2023Feb 13, 2024May 2, 2024Jul 20, 2024Oct 6, 2024Dec 24, 2024Mar 13, 2025

Evaluation results

MethodLinks
RAR-XLType=AR, Param.=955 M, Steps=2562025.031.5810.55306.98062—
RAR-LType=AR, Param.=461 M, Steps=2562025.031.712.086.37299.58160—
MAR-LType=Mask, Param.=479 M, Steps=1002025.031.781.272.322968160—
ARPG-XXLType=AR (Parallel), Param.=1.3 B, Steps=322025.031.8355.287.22336.18060—
ARPG-XXLType=AR (Parallel), Param.=1.3 B, Steps=642025.031.8631.657.18339.78159—
ARPG-XLType=AR (Parallel), Param.=719 M, Steps=322025.031.9380.564.65349.28061—
RAR-BType=AR, Param.=261 M, Steps=2562025.031.9514.124.65290.58258—
ARPG-XLType=AR (Parallel), Param.=719 M, Steps=642025.031.9945.094.57340.68061—
VAR-d24Type=VAR, Param.=1.0 B, Steps=102025.032.0949.9422.29312.98259—
RandAR-XXLType=AR (Parallel), Param.=1.4 B, Steps=882025.032.1510.4621.773227962—
RandAR-XLType=AR (Parallel), Param.=775 M, Steps=882025.032.2515.5113.52317.88060—
DiT-XL/2Type=Diffusion, Param.=675 M, Steps=2502025.032.270.912.14278.28357—
ARPG-LType=AR (Parallel), Param.=320 M, Steps=322025.032.3130.142.78297.78256—
MAR-BType=Mask, Param.=208 M, Steps=1002025.032.311.711.47281.78257—
PAR-XXLType=AR (Parallel), Param.=1.4 B, Steps=1472025.032.355.2326.25263.28062—
ARPG-LType=AR (Parallel), Param.=320 M, Steps=642025.032.3767.472.64293.78255—
VDM++params.=2B, steps=512, Classifier-free guidance=false2023.122.4——225.3———
RandAR-LType=AR (Parallel), Param.=343 M, Steps=882025.032.5525.37.32288.88158—
W.A.L.T-LCost (Iter x BS)=437kx1024, params.=460M, steps=50, Classifier-free guidance=false2023.122.56——215.1———
VAR-d20Type=VAR, Param.=600 M, Steps=102025.032.5775.3815.97302.68356—
NAR-XXLType=AR (Parallel), Param.=1.5 B, Steps=312025.032.5815.2326.25293.58257—
PAR-XLType=AR (Parallel), Param.=775 M, Steps=1472025.032.617.9117.13259.28062—
LlamaGen-XLType=AR, Param.=775 M, Steps=5762025.032.622.4617.11244.18057—
LlamaGen-XXLType=AR, Param.=1.4 B, Steps=5762025.032.621.5826.22244.18057—
NAR-XLType=AR (Parallel), Param.=816 M, Steps=312025.032.723.9717.13277.58158—
simple diffusionCost (Iter x BS)=500Kx2048, params.=2B, steps=512, Classifier-free guidance=false2023.122.77——211.8———
NAR-LType=AR (Parallel), Param.=372 M, Steps=312025.033.0642.7110.25263.98153—
LlamaGen-LType=AR, Param.=343 M, Steps=5762025.033.074.3310.23256.18352—
VAR-d16Type=VAR, Param.=310 M, Steps=102025.033.3123.2110.85274.48451—
RINCost (Iter x BS)=600kx1024, params.=410M, steps=1000, Classifier-free guidance=false2023.123.42——182———
PAR-LType=AR (Parallel), Param.=343 M, Steps=1472025.033.7614.7710.25218.98160—
DiT-L/2Type=Diffusion, Param.=458 M, Steps=2502025.035.021.321.62167.27557—
MaskDiTCost (Iter x BS)=1200kx1024, params.=736M, steps=40, Classifier-free guidance=false2023.125.69——178———
MaskGITType=Mask, Param.=227 M, Steps=82025.036.1846.181.71182.18051—
MDTCost (Iter x BS)=6500kx256, params.=676M, steps=250, Classifier-free guidance=false2023.126.23——143———
BigGAN-deepparams.=160M, steps=1, Classifier-free guidance=false2023.126.95——171.4———
ADMparams.=608M, steps=2000, Classifier-free guidance=false2023.127.49——127.5———
DiT-XL/2Cost (Iter x BS)=7000kx256, params.=675M, steps=250, Classifier-free guidance=false2023.129.62——121.5———
TFMQ-DMBits (W/A)=4/82023.1110.29——221.82——7.35
PTQDBits (W/A)=4/82023.1110.4——214.73——12.63
TFMQ-DMBits (W/A)=4/322023.1110.5——223.81——7.98
LDM-4Cost (Iter x BS)=178kx1200, params.=400M, steps=250, Classifier-free guidance=false2023.1210.56——103.5———
Q-DiffusionBits (W/A)=4/82023.1110.68——212.51——14.85
TFMQ-DMBits (W/A)=8/82023.1110.79——198.86——7.65
Full Prec.Bits (W/A)=32/322023.1110.91——235.64——7.67
PTQDBits (W/A)=4/322023.1111.65——201.78——9.06
Q-DiffusionBits (W/A)=4/322023.1111.87——213.56——8.76
PTQDBits (W/A)=8/82023.1111.94——153.92——8.03
PTQ4DMBits (W/A)=8/82023.1112.59——161.75———
Q-DiffusionBits (W/A)=8/82023.1112.8——187.65——9.87