Personalized Image Captioning on CapEval-QAs (test)
77.41-Concept Acc+CoViP
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CoViPBackbone=Qwen3-VL-8B, Post-training Method=LoRA, Category=Post-Training-based Personalized VLMs2026.02 | 77.4 | 94.8 | 68.4 | 94.1 | 65.2 | 94.8 | 59.7 | 92.8 | |
| Gemini-3.0-ProAccess=Close-sourced, Category=Proprietary VLMs2026.02 | 58.1 | 96.6 | 45.1 | 97.2 | 39 | 98.3 | 32.4 | 97.9 | |
| GPT-5Access=Close-sourced, Category=Proprietary VLMs2026.02 | 48.3 | 97.3 | 28.2 | 97.9 | 26.1 | 98.7 | 18.9 | 98.7 | |
| RePICBackbone=Qwen3-VL-8B, Post-training Method=LoRA, Category=Post-Training-based Personalized VLMs2026.02 | 44 | 97.1 | 31.7 | 97 | 29.2 | 97.8 | 24 | 97.2 | |
| Gemini-2.0-FlashAccess=Close-sourced, Category=Proprietary VLMs2026.02 | 41.9 | 96.7 | 28.6 | 97.3 | 26.6 | 98.3 | 23.1 | 98.3 | |
| Qwen3-VL-30B-A3BAccess=Open-sourced, Category=Open-Sourced VLMs2026.02 | 40.2 | 96.2 | 27.5 | 97.7 | 25.3 | 97.7 | 20.1 | 98.1 | |
| Qwen3-VL-8BAccess=Open-sourced, Category=Open-Sourced VLMs2026.02 | 39 | 97.5 | 25.6 | 97.7 | 23.3 | 98.1 | 18.6 | 98.1 | |
| GPT-4oAccess=Close-sourced, Category=Proprietary VLMs2026.02 | 34.2 | 98.2 | 21.6 | 98.6 | 20.4 | 99.3 | 15.3 | 99.2 | |
| RAPBackbone=Qwen3-VL-8B, Post-training Method=LoRA, Category=Post-Training-based Personalized VLMs2026.02 | 20.5 | 99 | 10.4 | 99.1 | 9.9 | 99.5 | 7.3 | 99.2 |