Streaming Video Understanding on RVS-Movie SVU (test)
53.1AccuracyProtoKV
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ProtoKVBackbone=Qwen2.5-VL-7B, Token budget |M|=24k2026.06 | 53.1 | 3.77 | |
| InfiniPot-VBackbone=Qwen2.5-VL-7B, Token budget |M|=24k2026.06 | 52.3 | 3.69 | |
| ProtoKVBackbone=LLaVA-OV-7B, Token budget |M|=24k2026.06 | 52.1 | 3.59 | |
| SWABackbone=Qwen2.5-VL-7B, Token budget |M|=24k2026.06 | 52 | 3.55 | |
| InfiniPot-VBackbone=LLaVA-OV-7B, Token budget |M|=24k2026.06 | 51.4 | 3.5 | |
| SWABackbone=LLaVA-OV-7B, Token budget |M|=24k2026.06 | 50.8 | 3.4 |