Text-to-Image Generation on GenEval 256x256
0.96Single Object CompositionDiffMoE-E16-T2I-Flow (w SFT)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DiffMoE-E16-T2I-Flow (w SFT)Avg. Act. Params.=1.2B, Total Params.=4.6B, SFT=true2025.03 | 0.96 | 0.53 | 0.46 | 0.78 | 0.13 | 0.2 | 0.51 | |
| Dense-DiT-T2I-Flow (w SFT)Avg. Act. Params.=1.2B, Total Params.=1.2B, SFT=true2025.03 | 0.93 | 0.48 | 0.5 | 0.73 | 0.07 | 0.2 | 0.49 | |
| DiffMoE-E16-T2I-Flow (w/o SFT)Avg. Act. Params.=1.2B, Total Params.=4.6B, SFT=false2025.03 | 0.9 | 0.42 | 0.24 | 0.59 | 0.19 | 0.28 | 0.44 | |
| Dense-DiT-T2I-Flow (w/o SFT)Avg. Act. Params.=1.2B, Total Params.=1.2B, SFT=false2025.03 | 0.8 | 0.33 | 0.24 | 0.53 | 0.1 | 0.26 | 0.38 |