Streaming Social Task Detection on MOSI, MOSEI, HateClipSeg Aggregated
1.03VLM Success RateStreamSense (Llama-3.2)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| StreamSense (Llama-3.2)Latency=0.3 s, GPU=29 GB2026.01 | 1.03 | 1.73 | |
| Qwen2.5Latency=0.7 s, GPU=20 GB2026.01 | 1 | 0 | |
| LLaVA-NextLatency=1.1 s, GPU=21 GB2026.01 | 1 | 0 | |
| Llama-3.2Latency=0.8 s, GPU=27 GB2026.01 | 1 | 0 | |
| OadTRLatency=0.1 s, GPU=2 GB2026.01 | 0 | 0 | |
| LSTRLatency=0.1 s, GPU=2 GB2026.01 | 0 | 0 | |
| Stream EncoderLatency=0.1 s, GPU=2 GB2026.01 | 0 | 0 |