Text-to-image generation on MS COCO 2017 (test)
21.88FIDMaskUNet
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MaskUNetzero-shot=true2025.05 | 21.88 | — | — | 0.33 | |
| PixArt-αMACs (T)=10.851, Resolution=512×512, Sampling steps=202026.04 | 21.95 | — | — | 0.164 | |
| DiffSparseMACs (T)=5.986, Resolution=512×512, Sampling steps=202026.04 | 22.42 | — | — | 0.165 | |
| ToCaMACs (T)=5.993, Resolution=512×512, Sampling steps=202026.04 | 23.02 | — | — | 0.165 | |
| LoRAzero-shot=true2025.05 | 23.18 | — | — | 0.33 | |
| SD 1.5zero-shot=true2025.05 | 23.39 | — | — | 0.33 | |
| Full Fine-tunezero-shot=true2025.05 | 24.45 | — | — | 0.33 | |
| 50% stepsMACs (T)=5.426, Resolution=512×512, Sampling steps=202026.04 | 25.05 | — | — | 0.163 | |
| FORA + TCCLatency (s)=0.5388, FLOPs=4.3469, Speed=1.82×, N=3, Calibration=TCC2026.05 | 27.35 | — | — | 0.3075 | |
| FlowTurboSample Config=H3P6R32024.09 | 27.6 | 137 | 5,386 | — | |
| ToCa + TCCLatency (s)=1.0480, FLOPs=5.1146, Speed=0.94×, N=3, R=90%, Calibration=TCC2026.05 | 27.79 | — | — | 0.3033 | |
| ToCaLatency (s)=0.9720, FLOPs=6.5357, Speed=1.01×, N=3, R=60%2026.05 | 28.03 | — | — | 0.3095 | |
| PixArt-αLatency (s)=0.9832, FLOPs=11.1225, Speed=1.00×2026.05 | 28.15 | — | — | 0.3103 | |
| FlowTurboSample Config=H1P6R32024.09 | 28.6 | 104.8 | 4,030 | — | |
| SDXLSample Config=15 steps DPM++ [23]2024.09 | 29.46 | 427.2 | 24,266 | — | |
| ToCaLatency (s)=0.9210, FLOPs=4.6580, Speed=1.07×, N=3, R=90%2026.05 | 29.63 | — | — | 0.3095 | |
| FORALatency (s)=0.4580, FLOPs=3.8902, Speed=2.15×, N=32026.05 | 29.83 | — | — | 0.3098 | |
| SD2.1Sample Config=15 steps DPM++ [23]2024.09 | 33.03 | 286 | 11,427 | — | |
| PixArt-σSample Config=15 steps DPM++ [23]2024.09 | 33.62 | 365.9 | 17,957 | — | |
| PixArt-α (50% steps)Latency (s)=0.4997, FLOPs=5.5613, Speed=1.97×, steps=50%2026.05 | 37.47 | — | — | 0.305 | |
| PixArt-αSample Config=15 steps DPM++ [23]2024.09 | 37.96 | 366.8 | 17,523 | — |