Fine-grained Image Captioning on DetailCaps (test)
72.2CAPTUREQwen2.5-VL+TDSR
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-VL+TDSRModel Architecture=Qwen2.5-VL, Model Scale=7B, Enhancement Method=TDSR2025.10 | 72.2 | 72.3 | 65.2 | 64.7 | |
| LLaVA-1.5+TDSRModel Architecture=LLaVA-1.5, Model Scale=7B, Enhancement Method=TDSR2025.10 | 66.7 | 66.2 | 62.4 | 63.4 | |
| Qwen2.5-VLModel Architecture=Qwen2.5-VL, Model Scale=7B, Enhancement Method=None2025.10 | 64.7 | 66.7 | 62.5 | 62.3 | |
| Mini-Gemini+OursBase Model=Mini-Gemini, Augmentation=Patch Matters / divide-then-aggregate2025.04 | 64.49 | 68.15 | 63.92 | 56.75 | |
| LLaVA1.6+OursBase Model=LLaVA-1.6, Augmentation=Patch Matters / divide-then-aggregate2025.04 | 63.51 | 66.95 | 63.15 | 55.8 | |
| FINECAPTIONModel Architecture=FINECAPTION, Model Scale=8B, Enhancement Method=None2025.10 | 63.4 | 63.7 | 58.1 | 58.3 | |
| VistaLLMModel Architecture=VistaLLM, Model Scale=13B, Enhancement Method=None2025.10 | 63.2 | 63.5 | 60.3 | 59.2 | |
| FerretModel Architecture=Ferret, Model Scale=13B, Enhancement Method=None2025.10 | 62.8 | 63.2 | 56.4 | 57.3 | |
| Mini-Gemini+ITBase Model=Mini-Gemini, Augmentation=IT2025.04 | 62.53 | 66.85 | 61.4 | 54.54 | |
| Mini-Gemini+SynBase Model=Mini-Gemini, Augmentation=Syn2025.04 | 62.53 | 66.68 | 61.49 | 54.77 | |
| Mini-GeminiBase Model=Mini-Gemini2025.04 | 62.37 | 66.77 | 61.15 | 54.4 | |
| LLaVA1.6+ITBase Model=LLaVA-1.6, Augmentation=IT2025.04 | 61.21 | 64.76 | 60.68 | 53.68 | |
| MiniGPT-v2Model Architecture=MiniGPT-v2, Model Scale=7B, Enhancement Method=None2025.10 | 61.2 | 62.7 | 55.9 | 56.8 | |
| LLaVA1.6+SynBase Model=LLaVA-1.6, Augmentation=Syn2025.04 | 61.16 | 65.17 | 60.24 | 53.44 | |
| LLaVA1.6Base Model=LLaVA-1.62025.04 | 61.03 | 64.59 | 60.39 | 53.73 | |
| ShikraModel Architecture=Shikra, Model Scale=13B, Enhancement Method=None2025.10 | 60.5 | 61.9 | 55.4 | 56.4 | |
| Alpha-CLIPModel Architecture=Alpha-CLIP, Model Scale=13B, Enhancement Method=None2025.10 | 59.2 | 61.9 | 57.2 | 56.5 | |
| KOSMOS-2Model Architecture=KOSMOS-2, Enhancement Method=None2025.10 | 58.5 | 60.8 | 53.1 | 54.2 | |
| LLaVA-1.5Model Architecture=LLaVA-1.5, Model Scale=7B, Enhancement Method=Patch Matters2025.10 | 58.05 | 62.2 | 56.1 | 52.5 | |
| LLaVA1.5+OursBase Model=LLaVA-1.5, Augmentation=Patch Matters / divide-then-aggregate2025.04 | 58.05 | 62.23 | 56.09 | 52.51 | |
| VisionLLM-HModel Architecture=VisionLLM-H, Model Scale=7B, Enhancement Method=None2025.10 | 57.9 | 59.3 | 53.4 | 53.9 | |
| LLaVA1.5+SynBase Model=LLaVA-1.5, Augmentation=Syn2025.04 | 57.51 | 62.06 | 55.54 | 51.07 | |
| LLaVA1.6+PoCaBase Model=LLaVA-1.6, Augmentation=PoCa2025.04 | 55.38 | 58.61 | 54.47 | 49.61 | |
| Mini-Gemini+PoCaBase Model=Mini-Gemini, Augmentation=PoCa2025.04 | 55.08 | 59.73 | 52.68 | 49.44 | |
| LLaVA-1.5Model Architecture=LLaVA-1.5, Model Scale=7B, Enhancement Method=IT2025.10 | 51.98 | 56.3 | 48.2 | 50.4 | |
| LLaVA1.5+ITBase Model=LLaVA-1.5, Augmentation=IT2025.04 | 51.98 | 56.37 | 48.22 | 50.44 | |
| LLaVA-1.5Model Architecture=LLaVA-1.5, Model Scale=7B, Enhancement Method=None2025.10 | 49.99 | 55.7 | 44.4 | 49.4 | |
| LLaVA1.5Base Model=LLaVA-1.52025.04 | 49.99 | 55.76 | 44.43 | 49.48 | |
| LLaVA1.5+PoCaBase Model=LLaVA-1.5, Augmentation=PoCa2025.04 | 47.38 | 54.57 | 40.64 | 46.28 |