Region-controlled text-to-image generation on COCO 30k 2014 (val)
36.8APReal Images
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Real Images2022.11 | 36.8 | 56.1 | 71.41 | — | — | |
| ReCoEvaluation Protocol=Fine-tune, Image Descr.=true, Region Descr.=true, Region Position=true2022.11 | 32 | 52.4 | 62.42 | 6.51 | 7.36 | |
| ReCo (Position Word)Evaluation Protocol=Fine-tune, Region Position=Text2022.11 | 2.3 | 7.5 | 42.02 | 15.54 | 8.82 | |
| ReCo (Region Descr.)Evaluation Protocol=Fine-tune, Region Descr.=true2022.11 | 1 | 2.9 | 32.32 | 24.88 | 9.11 | |
| ReCo (Image Descr.)Evaluation Protocol=Fine-tune, Image Descr.=true2022.11 | 0.9 | 2.6 | 29.12 | 27.78 | 10.44 | |
| SD V1.4Evaluation Protocol=Zero-shot, Image Descr.=true2022.11 | 0.7 | 2 | 26.75 | 35.8 | 13.4 | |
| SD V1.4Evaluation Protocol=Zero-shot, Image Descr.=true, Region Descr.=true2022.11 | 0.7 | 2 | 27.64 | 34.72 | 13.88 | |
| SD V1.4Evaluation Protocol=Zero-shot, Image Descr.=true, Region Position=Text2022.11 | 0.6 | 1.8 | 28.15 | 32.86 | 14.57 |