Text-to-Image Retrieval on MS-COCO (val test)
5.075Ave AesGPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4oAesthetic Condition=High, Relevance Condition=Medium, Controllable=No2026.02 | 5.075 | 0.352 | |
| Ours (QCQC)Aesthetic Condition=High, Relevance Condition=High, Controllable=Yes2026.02 | 5.024 | 0.387 | |
| Ours (QCQC)Aesthetic Condition=High, Relevance Condition=Low, Controllable=Yes2026.02 | 5.016 | 0.352 | |
| Ours (QCQC)Aesthetic Condition=High, Relevance Condition=Medium, Controllable=Yes2026.02 | 4.983 | 0.365 | |
| FTAesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.925 | 0.37 | |
| Ours (QCQC)Aesthetic Condition=Medium, Relevance Condition=Low, Controllable=Yes2026.02 | 4.911 | 0.354 | |
| LLaMA3Aesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.903 | 0.348 | |
| FT-CoCaAesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.878 | 0.346 | |
| Ours (QCQC)Aesthetic Condition=Medium, Relevance Condition=Medium, Controllable=Yes2026.02 | 4.873 | 0.37 | |
| Ours (QCQC)Aesthetic Condition=Medium, Relevance Condition=High, Controllable=Yes2026.02 | 4.862 | 0.387 | |
| FT-Blip2Aesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.828 | 0.35 | |
| PTAesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.819 | 0.343 | |
| PrefixAesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.817 | 0.349 | |
| Ours (QCQC)Aesthetic Condition=Low, Relevance Condition=Low, Controllable=Yes2026.02 | 4.811 | 0.356 | |
| Ours (QCQC)Aesthetic Condition=Low, Relevance Condition=Medium, Controllable=Yes2026.02 | 4.79 | 0.37 | |
| Ours (QCQC)Aesthetic Condition=Low, Relevance Condition=High, Controllable=Yes2026.02 | 4.773 | 0.382 | |
| GPT-4oAesthetic Condition=Low, Relevance Condition=Low, Controllable=No2026.02 | 4.673 | 0.371 |