Streaming Social Task Detection on MOSI
81.24AccuracyStreamSense (Llama-3.2)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| StreamSense (Llama-3.2)Latency=0.3 s, GPU=29 GB2026.01 | 81.24 | 79.26 | |
| Llama-3.2Latency=0.8 s, GPU=27 GB2026.01 | 80.85 | 79.05 | |
| LLaVA-NextLatency=1.1 s, GPU=21 GB2026.01 | 80.1 | 78.42 | |
| Qwen2.5Latency=0.7 s, GPU=20 GB2026.01 | 79.8 | 77.71 | |
| Stream EncoderLatency=0.1 s, GPU=2 GB2026.01 | 71.72 | 70.79 | |
| OadTRLatency=0.1 s, GPU=2 GB2026.01 | 71.1 | 68.73 | |
| LSTRLatency=0.1 s, GPU=2 GB2026.01 | 65.56 | 64.55 |