Video Grounding on E.T. Bench-Grounding (test)
52TVG F1E.M.Ground
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| E.M.GroundEvaluation Protocol=Zero-shot, Model Size=3.8B2026.02 | 52 | 14 | 30.9 | 32.8 | — | — | |
| Gemini-1.5-FlashModel Type=Proprietary, Tested on Subset=true2025.01 | 43.9 | 5.4 | 27 | 5.4 | 60.8 | 28.5 | |
| Gemini-1.5-ProModel Type=Proprietary, Tested on Subset=true2025.01 | 43.1 | 6.2 | 33.8 | 7.9 | 47 | 27.6 | |
| GPT-4oModel Type=Proprietary, Tested on Subset=true2025.01 | 40.4 | 4.5 | 20 | 17.6 | 56.9 | 27.9 | |
| Qwen2-VL-7BModel Type=Open-source, Parameter Scale=7B, Fine-tuned on E.T. Instruct 164K=true2025.01 | 39.7 | 7 | 26.9 | 17.1 | 66.9 | 33.5 | |
| Tarsier-7BModel Type=Open-source, Parameter Scale=7B, Fine-tuned on E.T. Instruct 164K=true2025.01 | 39.6 | 9 | 25 | 25.4 | 47.6 | 30.9 | |
| E.T. ChatModel Type=Open-source, Parameter Scale=<10B, Fine-tuned on E.T. Instruct 164K=true2025.01 | 38.6 | 10.2 | 30.8 | 25.4 | 62.5 | 33.5 | |
| E.T.ChatEvaluation Protocol=Zero-shot, Model Size=3.8B2026.02 | 38.6 | 10.2 | 30.8 | 25.4 | — | — | |
| Tarsier2-7BModel Type=Open-source, Parameter Scale=7B, Fine-tuned on E.T. Instruct 164K=true2025.01 | 38.4 | 11 | 31.8 | 19.4 | 66.8 | 35.5 | |
| GPT-4VModel Type=Proprietary, Tested on Subset=true2025.01 | 27 | 1.8 | 18 | 28.6 | 55.1 | 26.1 | |
| LITAEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 26.2 | 3.9 | 10.1 | 29.1 | — | — | |
| LITAModel Type=Open-source, Parameter Scale=<10B, Tested on Subset=true2025.01 | 22.2 | 4.6 | 18 | 29.7 | 23.9 | 19.7 | |
| VTG-LLMModel Type=Open-source, Parameter Scale=<10B, Tested on Subset=true2025.01 | 15.9 | 3.7 | 14.4 | 26.8 | 48.2 | 21.8 | |
| VTG-LLMEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 15.9 | 3.7 | 14.4 | 26.8 | — | — | |
| VTimeLLMEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 7.6 | 1.9 | 18.2 | 15.9 | — | — | |
| Video-LLaVAEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 7 | 1.9 | 15 | 0.3 | — | — | |
| Video-ChatGPTEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 7 | 1.3 | 15.1 | 8.4 | — | — | |
| PLLaVAEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 6.9 | 1.1 | 5.7 | 0.3 | — | — | |
| LLaMA-VIDEvaluation Protocol=Zero-shot, Model Size=7B2026.02 | 5.5 | 1.2 | 8 | 1.4 | — | — | |
| TimeChatModel Type=Open-source, Parameter Scale=<10B, Fine-tuned on E.T. Instruct 164K=true2025.01 | — | — | — | — | — | 24.3 |