Multi-Choice on Extreme Illumination Triplet Instruct-Following Dataset
53.13AccuracyOurs-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ours-7BMLLM Type=Event-Enhanced Model, Backbone=Qwen-7B2026.03 | 53.13 | 85.43 | |
| Baseline-7BMLLM Type=Event-Enhanced Model, Backbone=Qwen-7B2026.03 | 44.71 | 81.95 | |
| Ours-3BMLLM Type=Event-Enhanced Model, Backbone=Qwen-3B2026.03 | 34.48 | 80.35 | |
| InternVLMLLM Type=General Model2026.03 | 31.5 | 75.46 | |
| Baseline-3BMLLM Type=Event-Enhanced Model, Backbone=Qwen-3B2026.03 | 29.86 | 77.04 | |
| LLaVA-7BMLLM Type=General Model2026.03 | 18.43 | 65.39 | |
| Q-InstructMLLM Type=Illumination-aware Model2026.03 | 15.86 | 65.09 | |
| Deepseek-VLMLLM Type=General Model2026.03 | 7 | 55.7 | |
| EventGPTMLLM Type=Event-specialized Model2026.03 | 3.16 | 73.6 |