Text-to-Image Generation on T2I-CompBench (Color, Shape, Texture Fidelity)
77.07Color FidelityUniTok
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| UniTokLLM=LLaMA-2-7B, PT Data Size=70M2025.05 | 77.07 | 52.98 | 65.21 | |
| DG+CFG (Ours)Model=SD32026.07 | 59.76 | 58.81 | 58.07 | |
| S²-GuidanceModel=SD32026.07 | 59.63 | 58.71 | 56.77 | |
| Slot-MLLM (14B)LLM=Qwen2.5-14B-Instruct, PT Data Size=25M2025.05 | 58.85 | 40.25 | 52.74 | |
| DG (Ours)Model=SD32026.07 | 58.62 | 58.06 | 57.97 | |
| VILA-ULLM=LLaMA-2-7B, PT Data Size=23M2025.05 | 53.89 | 41.23 | 50.79 | |
| CFGModel=SD32026.07 | 53.61 | 51.2 | 52.45 | |
| Slot-MLLM (7B)LLM=Vicuna-7B, PT Data Size=25M2025.05 | 53.08 | 35.89 | 46.16 | |
| CFG-ZeroModel=SD32026.07 | 52.7 | 52.84 | 53.37 | |
| CFG++Model=SD32026.07 | 46.39 | 47.18 | 46.33 | |
| APGModel=SD32026.07 | 45.28 | 46.27 | 46.84 | |
| LaVITLLM=LLaMA-7B, PT Data Size=100M2025.05 | 27.3 | 31.91 | 39.88 | |
| SEED-LLaMALLM=Vicuna-7B, PT Data Size=77M2025.05 | 17.41 | 20.33 | 23.61 |