Text-to-Image Generation on Multi-concept datasets five composition pairs (test)
86.65Text AlignmentCustom Diffusion
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Custom Diffusiontraining_strategy=Joint, baseline_comparison=Textual Inversion2022.12 | 86.65 | 81.89 | |
| Custom Diffusiontraining_strategy=Optimization, baseline_comparison=Textual Inversion2022.12 | 81.22 | 83.11 | |
| Custom Diffusiontraining_strategy=Joint, baseline_comparison=Ours (w/ fine-tune all)2022.12 | 59 | 59.12 | |
| Custom Diffusiontraining_strategy=Optimization, baseline_comparison=Ours (w/ fine-tune all)2022.12 | 57.6 | 53.49 | |
| Custom Diffusiontraining_strategy=Optimization, baseline_comparison=DreamBooth2022.12 | 57 | 61.75 | |
| Custom Diffusiontraining_strategy=Joint, baseline_comparison=DreamBooth2022.12 | 56.39 | 61.8 |