Research map / Video / Video-language understanding
Decouple and Cache: KV Cache Construction for Streaming Video Understanding
Zhanzhong Pang, Dibyadip Chatterjee, Fadime Sener, Angela Yao
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 |
|---|---|---|---|
| add decoupled instant cache (DSCache) over uniform cache | LLaVA-OV-7B · StreamingBench Real-Time Visual Understanding | accuracy: 76.92 -> 79.12 | Improved |
| add decoupled instant cache (DSCache) over uniform cache | Qwen2.5-VL-7B · StreamingBench Real-Time Visual Understanding | accuracy: 78.56 -> 82.32 | Improved |
| add decoupled instant cache (DSCache) over uniform cache | LLaVA-OV-7B · OVO-Bench | overall accuracy: 55.6 -> 57.5 | Improved |
| add decoupled instant cache (DSCache) over uniform cache | Qwen2.5-VL-7B · OVO-Bench | overall accuracy: 52.2 -> 55.4 | Improved |
| add DSCache over uniform cache | LLaVA-OV-7B · RVS-Movie | accuracy: 48.1 -> 49.4 | Improved |
| add DSCache over uniform cache | LLaVA-OV-7B · StreamingBench subtasks CR and CT | accuracy | Hurt |
| include attention sink in instant cache construction (ablation) | DSCache LLaVA-OV-7B · StreamingBench Real-Time Visual Understanding | accuracy: 74.27 -> 79.12 | Improved |
| include attention sink in instant cache construction (ablation) | DSCache Qwen2.5-VL-7B · StreamingBench Real-Time Visual Understanding | accuracy: 76.31 -> 82.32 | Improved |
| include attention sink in cumulative cache construction (ablation) | DSCache LLaVA-OV-7B · StreamingBench Real-Time Visual Understanding | accuracy: 79.06 -> 79.12 | No clear effect |
| include attention sink in cumulative cache construction (ablation) | DSCache Qwen2.5-VL-7B · OVO-Bench | overall accuracy: 53.41 -> 55.40 | Improved |
| combine DSCache with ReKV (ablation) | DSCache LLaVA-OV-7B · OVO-Bench | overall accuracy: 57.5 -> 58.5 | Improved |
| combine DSCache with InfiniPot-V (ablation) | DSCache LLaVA-OV-7B · StreamingBench Real-Time Visual Understanding | accuracy: 79.12 -> 79.60 | Improved |
Closest papers by meaning
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.89)
- InfiniPot-V: Memory-Constrained KV Cache Compression for Streaming Video Understanding (NeurIPS 2025 · similarity 0.88)
- Streaming Video Question-Answering with In-context Video KV-Cache Retrieval (ICLR 2025 · similarity 0.87)
- VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion (NeurIPS 2026 · similarity 0.87)
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.87)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.86)
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.
