Conditional Image Generation, Editing, and Composition on ImagenHub (test)
0.493Subject-driven IG ScoreCIGEVAL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CIGEVALBackbone=GPT-4o2025.04 | 0.493 | 0.5027 | 0.5465 | 0.409 | 0.5402 | 0.5185 | 0.4931 | 0.4625 | |
| VIESCOREBackbone=GPT-4o2025.04 | 0.4806 | 0.4989 | 0.5421 | 0.4062 | 0.4972 | 0.48 | 0.4516 | 0.4459 | |
| Human Raters2025.04 | 0.478 | 0.5044 | 0.539 | 0.423 | 0.5443 | 0.4887 | 0.5927 | 0.47 | |
| CIGEVALBackbone=Qwen2.5-VL-7B-Instruct2025.04 | 0.4374 | 0.4347 | 0.4685 | 0.2567 | 0.3752 | 0.4863 | 0.3251 | 0.378 | |
| DINO2025.04 | 0.416 | — | — | — | — | 0.3022 | 0.0979 | — | |
| CLIP-I2025.04 | 0.2961 | — | — | — | — | 0.2834 | 0.1512 | — | |
| CIGEVALBackbone=LLaMA3-LLaVA-NeXT-8B2025.04 | 0.2891 | 0.142 | 0.2843 | 0.0744 | 0.4487 | -0.0699 | 0.3704 | 0.2164 | |
| VIESCOREBackbone=LLaMA3-LLaVA-NeXT-8B2025.04 | 0.1592 | 0.1948 | 0.2037 | 0.0363 | 0.4001 | -0.1153 | 0.1308 | 0.1432 | |
| VIESCOREBackbone=Qwen2.5-VL-7B-Instruct2025.04 | 0.1261 | 0.4218 | 0.3555 | 0.0252 | 0.2836 | -0.0452 | 0.3328 | 0.2516 | |
| CLIPScore2025.04 | — | 0.0817 | — | — | — | — | — | — | |
| LPIPS2025.04 | — | — | -0.1012 | 0.0956 | 0.3699 | — | — | — |