Text-to-Image on GenEval (DINO, VS, DreamSim, MSS, CLIPScore)
0.7DINOGroupInference
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GroupInferenceBase Model=FLUX.1-dev, Latency=4.55s, Selected Samples=8, Candidate Pool Size=1282026.06 | 0.7 | 3.61 | 0.37 | 0.19 | 32.37 | |
| Feature Self-GuidanceBase Model=FLUX.1-dev, Latency=1.70s2026.06 | 0.68 | 3.57 | 0.4 | 0.19 | 32.05 | |
| GroupInferenceBase Model=FLUX.1-dev, Latency=2.86s, Selected Samples=8, Candidate Pool Size=642026.06 | 0.67 | 3.37 | 0.35 | 0.21 | 32.49 | |
| CNOBase Model=FLUX.1-dev, Latency=1.75s2026.06 | 0.63 | 3.46 | 0.39 | 0.22 | 31.97 | |
| Shielded DiffusionBase Model=FLUX.1-dev, Latency=1.59s2026.06 | 0.61 | 3.35 | 0.4 | 0.23 | 32.11 | |
| Particle GuidanceBase Model=FLUX.1-dev, Latency=2.94s2026.06 | 0.58 | 3.16 | 0.31 | 0.29 | 32.22 | |
| IIDBase Model=FLUX.1-dev, Latency=1.59s2026.06 | 0.57 | 2.77 | 0.27 | 0.32 | 32.43 | |
| Interval GuidanceBase Model=FLUX.1-dev, Latency=1.59s2026.06 | 0.57 | 2.88 | 0.28 | 0.3 | 32.46 |