List your company
ResearchFormerly WizWand

Class-to-image generation on ImageNet

2.31FIDDDPM
2.13923.29214.4455.5979Dec 14, 2025Jan 9, 2026Feb 4, 2026Mar 2, 2026Mar 28, 2026Apr 23, 2026May 20, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DDPMBackbone=DiT-XL/2, Sampling steps=250, Latency(s)=49.564, FLOPs(T)=118.682025.122.314.988258243.42
DDIMBackbone=DiT-XL/2, Sampling steps=50, Latency(s)=1.564, Speed=1.00×2025.122.324.32——241.25
X-Slim(S)-slowBackbone=DiT-XL/2, Mode=slow, Sampling steps=25, Latency(s)=0.668, Speed=2.34×2025.122.324.34——242.11
X-Slim(C2F)-slowBackbone=DiT-XL/2, Mode=slow, Sampling steps=25, Latency(s)=0.611, Speed=2.56×2025.122.334.49——237.97
TaylorSeerBackbone=DiT-XL/2, N=3, Sampling steps=25, Latency(s)=0.764, Speed=2.05×2025.122.344.69——238.07
X-Slim(S)-fastBackbone=DiT-XL/2, Mode=fast, Sampling steps=20, Latency(s)=0.553, Speed=2.83×2025.122.384.55——236.3
X-Slim(C2F)-fastBackbone=DiT-XL/2, Mode=fast, Sampling steps=20, Latency(s)=0.5, Speed=3.13×2025.122.424.59——235.64
DDIMBackbone=DiT-XL/2, Sampling steps=50, Latency(s)=4.549, FLOPs(T)=23.74, Speed=1.00x2025.122.434.48059241.25
LineARρ=1/4, Backbone=LlamaGen-XL2026.052.64———279.95
TaylorSeerBackbone=DiT-XL/2, N=5, Sampling steps=20, Latency(s)=0.663, Speed=2.36×2025.122.655.34——231.11
LineARρ=1/6, Backbone=LlamaGen-XL2026.052.68———277.04
∆-DiTBackbone=DiT-XL/2, N=2, Sampling steps=50, Latency(s)=0.875, Speed=1.79×2025.122.694.67——225.99
HeadKVρ=1/4, Backbone=LlamaGen-XL2026.052.69———276.7
HeadKVρ=1/6, Backbone=LlamaGen-XL2026.052.71———276.43
HeadKVρ=1/8, Backbone=LlamaGen-XL2026.052.75———273.14
Fullρ=Full, Backbone=LlamaGen-XL2026.052.77———282.41
ProCacheBackbone=DiT-XL/2, Activation budget (B)=21, Latency(s)=2.049, FLOPs(T)=8.66, Speed=2.74x2025.122.934.948157235.4
LineARρ=1/8, Backbone=LlamaGen-XL2026.052.95———282.91
ProCacheBackbone=DiT-XL/2, Activation budget (B)=17, Latency(s)=1.725, FLOPs(T)=8.18, Speed=2.90x2025.122.964.938057232.85
ToCaBackbone=DiT-XL/2, N=3, Latency(s)=2.087, FLOPs(T)=10.23, Speed=2.32x2025.123.045.147956230.7
DDIMBackbone=DiT-XL/2, Sampling steps=25, Latency(s)=2.263, FLOPs(T)=11.87, Speed=2.00x2025.123.184.747958232.01
DDIMBackbone=DiT-XL/2, Sampling steps=25, Latency(s)=0.818, Speed=1.91×2025.123.184.74——232.01
FORABackbone=DiT-XL/2, N=3, Sampling steps=25, Latency(s)=0.701, Speed=2.23×2025.123.556.36——229.02
ProCacheBackbone=DiT-XL/2, Activation budget (B)=13, Latency(s)=1.654, FLOPs(T)=6.42, Speed=3.69x2025.123.596.058055225.23
ToCaBackbone=DiT-XL/2, N=4, Latency(s)=2.063, FLOPs(T)=8.73, Speed=2.72x2025.123.645.557856223.25
Δ-DiTBackbone=DiT-XL/2, N=3, Latency(s)=2.572, FLOPs(T)=16.46, Speed=1.47x2025.123.755.77754207.57
∆-DiTBackbone=DiT-XL/2, N=3, Sampling steps=50, Latency(s)=0.615, Speed=2.54×2025.123.755.7——207.57
DDIMBackbone=DiT-XL/2, Sampling steps=20, Latency(s)=1.862, FLOPs(T)=9.49, Speed=2.50x2025.123.815.157858221.43
DDIMBackbone=DiT-XL/2, Sampling steps=20, Latency(s)=0.679, Speed=2.30×2025.123.815.15——221.43
FORABackbone=DiT-XL/2, N=3, Latency(s)=2.191, FLOPs(T)=8.59, Speed=2.76x2025.123.886.437954229.02
DDIMBackbone=DiT-XL/2, Sampling steps=17, Latency(s)=1.563, FLOPs(T)=8.07, Speed=2.94x2025.124.585.767756208.72
FORABackbone=DiT-XL/2, N=5, Sampling steps=20, Latency(s)=0.602, Speed=2.95×2025.126.5811.29——193.01