Region-level instance understanding on ViP-Bench Synthesized visual prompts
58.9Recognition AccuracyLLaVA-NeXT-INST-IT-Qwen2-7B
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| LLaVA-NeXT-INST-IT-Qwen2-7BBase Model=Qwen2-7B, Visual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 58.9 | 24.5 | 48.5 | 12.9 | 48.2 | 46.3 | 50.5 | |
| GPT-4V-turbo-detail:highDetail Level=high, Visual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 58.1 | 69.8 | 59.5 | 71 | 61.4 | 51.9 | 60.7 | |
| ViP-LLaVA-7BVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 54.8 | 18.8 | 52.9 | 9.7 | 53.9 | 42.5 | 45.5 | |
| GPT-4V-turbo-detail:lowDetail Level=low, Visual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 53.2 | 50.3 | 55.6 | 67.7 | 57.5 | 57.5 | 52.8 | |
| LLaVA-NeXT-INST-IT-Vicuna-7BBase Model=Vicuna-7B, Visual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 51.3 | 23.7 | 54.2 | 12.9 | 64.3 | 46.2 | 45.1 | |
| LLAVA-1.5-7BVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 50.8 | 12.4 | 49.2 | 6.5 | 51.8 | 23.8 | 41.6 | |
| Qwen-VL-ChatVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 43 | 30.4 | 40.2 | 9.7 | 25.7 | 28.7 | 39.2 | |
| Shikra-7BVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 40.2 | 10 | 28 | 3.5 | 18.9 | 20.6 | 33.7 | |
| InstructBLIP-7BVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 36.9 | 16.3 | 34.2 | 22.3 | 26.8 | 7.5 | 31.7 | |
| GPT4ROI-7BVisual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 35.6 | 16.7 | 29.7 | 9.7 | 32.5 | 13.8 | 35.1 | |
| Kosmos-2Visual prompt source=Synthesized, Evaluation protocol=Zero-shot2024.12 | 29.5 | 14.2 | 18.5 | 9.7 | 7.5 | 21.9 | 26.9 |