Video Question Answering on ForeSeaQA
65.4AccuracyForeSea
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ForeSeaRetrieval latency (s)=0.5, Generation latency (s)=2.1, TTFT (s)=1.7, Total latency (s)=2.62026.03 | 65.4 | 13.8 | |
| ForeSea-GlobalRetrieval latency (s)=0.5, Generation latency (s)=0.9, TTFT (s)=0.6, Total latency (s)=1.42026.03 | 65.1 | 16.1 | |
| VideoRAGBackbone=LLaVA-Video-7B-Qwen2, Retrieval latency (s)=2.4, Generation latency (s)=2.8, TTFT (s)=2.3, Total latency (s)=5.22026.03 | 61.9 | 2.8 | |
| VideoLLaMA3-7BRetrieval latency (s)=0.0, Generation latency (s)=3.8, TTFT (s)=3.6, Total latency (s)=3.82026.03 | 61.6 | 10.9 | |
| Qwen2.5-VL-7B-InstructRetrieval latency (s)=0.0, Generation latency (s)=2.1, TTFT (s)=1.7, Total latency (s)=2.12026.03 | 58.9 | 8.1 | |
| T*Backbone=Qwen2.5-VL-7B-Instruct, Retrieval latency (s)=6.8, Generation latency (s)=0.9, TTFT (s)=0.6, Total latency (s)=7.62026.03 | 41.1 | 4.9 |