Text-to-Image Generation on GenEval (test) (Perceptual Metrics Suite)
0.14Latency (s)IID
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| IIDBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x5122026.06 | 0.14 | 0.46 | 2.14 | 0.19 | 0.41 | 33.78 | 0.31 | 23.5 | |
| CADSBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x5122026.06 | 0.14 | 0.5 | 2.42 | 0.09 | 0.51 | 19.01 | 0.13 | 17.27 | |
| Shielded DiffusionBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x512, Repellence radius r=160, Overcompensation factor λ=1.62026.06 | 0.15 | 0.55 | 2.45 | 0.23 | 0.33 | 33.77 | 0.3 | 23.19 | |
| Feature Self-GuidanceBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x5122026.06 | 0.16 | 0.6 | 2.73 | 0.26 | 0.27 | 33.27 | 0.3 | 23.09 | |
| CNOBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x512, Optimization steps=3, gamma=1.0, window size w=16, learning rate eta=0.012026.06 | 0.23 | 0.55 | 2.42 | 0.24 | 0.32 | 33.79 | 0.3 | 23.43 | |
| Particle GuidanceBase Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x512, Coefficient=150, Feature Extractor=DINO-VIT-s/82026.06 | 0.45 | 0.48 | 2.36 | 0.22 | 0.36 | 33.71 | 0.3 | 23.37 | |
| GroupInference (8 of 64)Base Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x512, Candidate samples=64, Selected samples=8, Pruning ratio=0.52026.06 | 1.14 | 0.48 | 2.35 | 0.18 | 0.46 | 33.54 | 0.28 | 22.82 | |
| GroupInference (8 of 128)Base Model=FLUX.2-Klein-4B, Inference timesteps=4, Guidance scale=0, Resolution=512x512, Candidate samples=128, Selected samples=8, Pruning ratio=0.52026.06 | 1.64 | 0.51 | 2.52 | 0.2 | 0.44 | 33.51 | 0.28 | 22.74 |