Streaming Social Task Detection on MOSEI
85.64AccuracyStreamSense (Llama-3.2)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| StreamSense (Llama-3.2)Latency=0.3 s, GPU=29 GB2026.01 | 85.64 | 74.23 | |
| Llama-3.2Latency=0.8 s, GPU=27 GB2026.01 | 85.35 | 73.43 | |
| Qwen2.5Latency=0.7 s, GPU=20 GB2026.01 | 84.44 | 72.35 | |
| LLaVA-NextLatency=1.1 s, GPU=21 GB2026.01 | 83.77 | 71.87 | |
| Stream EncoderLatency=0.1 s, GPU=2 GB2026.01 | 76.52 | 65.7 | |
| OadTRLatency=0.1 s, GPU=2 GB2026.01 | 73.86 | 64.1 | |
| LSTRLatency=0.1 s, GPU=2 GB2026.01 | 68.97 | 60.82 |