Text-to-Image In-Context Learning on CoBSAT
59.5Example Preference ProportionToT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ToTReasoning Strategy=Tree-of-Thoughts, Reasoning Backbone=SEED-LLaMA, Image Generator=Stable Diffusion, Fine-tuning=None, Reasoning Depth=4, Pruning Threshold=0.082026.07 | 59.5 | 68.1 | 65.5 | |
| BaselineReasoning Strategy=Baseline (direct prompt construction), Reasoning Backbone=SEED-LLaMA, Image Generator=Stable Diffusion, Fine-tuning=None2026.07 | 21.7 | 12.1 | 15.7 | |
| CoTReasoning Strategy=Chain-of-Thought (single linear trace), Reasoning Backbone=SEED-LLaMA, Image Generator=Stable Diffusion, Fine-tuning=None2026.07 | 18.8 | 19.8 | 18.8 |