Temporal Search on LVBench Haystack
6.3Temporal PrecisionReflect-R1
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Reflect-R1Base Model=Qwen2.5VL-7B, # Frame=10.18†, Search Strategy=Active Tool-Augmented Search2026.06 | 6.3 | 21.9 | 8.9 | 63.8 | 76.1 | 69.5 | |
| TimeSearch-RBase Model=Qwen2.5VL-7B, # Frame=9.32†, Search Strategy=Active Tool-Augmented Search2026.06 | 5.5 | 21.2 | 8.1 | 63.2 | 76.5 | 69.2 | |
| T*Base Model=GPT-4o, # Frame=32, Search Strategy=Adaptive Temporal Search2026.06 | 1.7 | 28.2 | 3.1 | 58.3 | 83.2 | 67.8 | |
| T*Base Model=GPT-4o, # Frame=8, Search Strategy=Adaptive Temporal Search2026.06 | 1.6 | 7.1 | 2.5 | 58.4 | 72.7 | 64.3 | |
| Retrieval-basedBase Model=GPT-4o, # Frame=8, Search Strategy=Adaptive Temporal Search2026.06 | 1.5 | 6.3 | 2.3 | 63.1 | 65.5 | 64.1 | |
| UniformBase Model=Qwen2.5VL-7B, # Frame=8, Search Strategy=Static Frame Sampling2026.06 | 1.4 | 6.3 | 2.2 | 56 | 72 | 62.7 | |
| UniformBase Model=GPT-4o, # Frame=8, Search Strategy=Static Frame Sampling2026.06 | 1.4 | 6.3 | 2.2 | 56 | 72 | 62.7 | |
| UniformBase Model=GPT-4o, # Frame=32, Search Strategy=Static Frame Sampling2026.06 | 1.4 | 24.9 | 2.7 | 58.7 | 81.6 | 67.3 | |
| Retrieval-basedBase Model=GPT-4o, # Frame=32, Search Strategy=Adaptive Temporal Search2026.06 | 1.3 | 21.8 | 2.4 | 59.9 | 80.8 | 67.8 | |
| VideoAgentBase Model=GPT-4, # Frame=10.1†, Search Strategy=Adaptive Temporal Search2026.06 | 1.2 | 8.5 | 2.1 | 58.8 | 73.2 | 64.7 |