Research map / Video / Video-language understanding
ProtoKV: Streaming Video Understanding under Delayed Query with Summary-State Memory
Le Tu Ngoc Minh, Jinyeong Lim, Dongsu Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| summarize far history as prototype bank instead of retaining tokens (ProtoKV vs InfiniPot-V) | LLaVA-OV-7B · RVS-Ego | accuracy (%): 57.8 -> 58.6 | Improved |
| ProtoKV vs InfiniPot-V on StreamingBench real-time | Qwen2.5-VL-7B · StreamingBench | accuracy (%): 76.4 -> 77.3 | Improved |
| add far prototype memory to near window (ablation) | ProtoKV · RVS-Ego at delay 30 min | accuracy (%): 40.5 -> 54.3 | Improved |
| add exact near window to far-only memory (ablation) | ProtoKV · RVS-Ego at delay 0 | accuracy (%): 54.5 -> 59.0 | Improved |
| add residual statistics (ablation) | ProtoKV · RVS-Ego at delay 30 min | accuracy (%): 51.0 -> 54.3 | Improved |
| add mass-aware log bias (ablation) | ProtoKV · RVS-Ego at delay 30 min | accuracy (%): 51.0 -> 54.3 | Improved |
| add continuity-aware assignment (spatial term) (ablation) | ProtoKV · RVS-Ego at delay 30 min | accuracy (%): 47.5 -> 54.3 | Improved |
| add residual statistics at no delay (ablation) | ProtoKV · RVS-Ego at delay 0 | accuracy (%): 58.5 -> 59.0 | No clear effect |
| ProtoKV vs SWA/InfiniPot-V at long query delay | ProtoKV · RVS-Movie, largest delays | accuracy gain (pp): +20.4pp over SWA, +12.5pp over token retention | Improved |
| ProtoKV vs InfiniPot-V on offline MLVU at 3K budget | Qwen2-VL-7B · MLVU | accuracy (%): 67.2 -> 67.0 | No clear effect |
| ProtoKV vs InfiniPot-V on offline VideoMME at 3K budget | Qwen2-VL-7B · VideoMME | accuracy (%): 61.2 -> 62.1 | Improved |
| replace ReKV offloading with GPU-resident ProtoKV | LLaVA-OV-7B · RVS-Ego | TTFT (s): 1.43 -> 1.06 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- InfiniPot-V — InfiniPot-V: Memory-Constrained KV Cache Compression for Streaming Video Understanding (NeurIPS 2025) · better on 10 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.89)
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.89)
- ReWind: Understanding Long Videos with Instructed Learnable Memory (CVPR 2025 · similarity 0.88)
- InfiniPot-V: Memory-Constrained KV Cache Compression for Streaming Video Understanding (NeurIPS 2025 · similarity 0.88)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.88)
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
