Perception on EventDrive
85.44Accuracy @ DEventDrive-VLM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EventDrive-VLMInput Modality=Event + Frame, Evaluation Protocol=Fine-tuned2026.06 | 85.44 | 86.64 | 78.89 | |
| Qwen2.5-VL-7B-Instruct*Input Modality=Frame, Evaluation Protocol=Fine-tuned2026.06 | 81.52 | 84.69 | 75.88 | |
| LLaVA-OneVision-1.5-8BInput Modality=Frame, Evaluation Protocol=Zero-shot2026.06 | 80.81 | 86.35 | 72.42 | |
| InternVL3-8BInput Modality=Frame, Evaluation Protocol=Zero-shot2026.06 | 78.61 | 86.27 | 74.37 | |
| InternVL2.5-8BInput Modality=Frame, Evaluation Protocol=Zero-shot2026.06 | 78.42 | 83.41 | 75.86 | |
| Qwen2.5-VL-7B-InstructInput Modality=Frame, Evaluation Protocol=Zero-shot2026.06 | 77.71 | 76.92 | 62.46 | |
| EventDrive-VLMInput Modality=Event, Evaluation Protocol=Fine-tuned2026.06 | 62.51 | 67.07 | 62.39 | |
| LLaVA-v1.6-Mistral-7B-hfInput Modality=Frame, Evaluation Protocol=Zero-shot2026.06 | 58.65 | 59.24 | 50.05 | |
| EventGPT-7BInput Modality=Event, Evaluation Protocol=Zero-shot2026.06 | 51.4 | 54.97 | 52.25 |