Real-time visual understanding on StreamingBench SVU (test)
77.3AccuracyProtoKV
Evaluation Results
| Method | Links | |
|---|---|---|
| ProtoKVBackbone=Qwen2.5-VL-7B, Token budget |M|=4k2026.06 | 77.3 | |
| InfiniPot-VBackbone=Qwen2.5-VL-7B, Token budget |M|=4k2026.06 | 76.4 | |
| ProtoKVBackbone=LLaVA-OV-7B, Token budget |M|=4k2026.06 | 76.3 | |
| SWABackbone=Qwen2.5-VL-7B, Token budget |M|=4k2026.06 | 75.9 | |
| InfiniPot-VBackbone=LLaVA-OV-7B, Token budget |M|=4k2026.06 | 75.2 | |
| SWABackbone=LLaVA-OV-7B, Token budget |M|=4k2026.06 | 72.9 |