Research map / Video / Video-language understanding
Think in Cloud, Look at Edges: Semantic-Driven Query Decomposition for Efficient Video Reasoning
Wenhao Zou, Zhijie Cai, Minchen Yu, Zongshuai Zhang, Guangxu Zhu
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 cloud query decomposition (vs flat Top-K) (ablation) | SCOPE 3B · Video-MME K=16 | accuracy: 55.04 -> 56.08 | Improved |
| add semantic-driven budget allocation (vs random) (ablation) | SCOPE 3B · Video-MME K=16 | accuracy: 56.08 -> 56.32 | Improved |
| add semantic-driven budget allocation (vs random) (ablation) | SCOPE 3B · Video-MME K=32 | accuracy: 57.17 -> 59.33 | Improved |
| add semantic-driven budget allocation (vs random) (ablation) | SCOPE 3B · Video-MME K=64 | accuracy: 59.78 -> 60.07 | No clear effect |
| add query decomposition (w/ Decomp. vs w/o) (ablation) | Top-K selection, Qwen2.5-VL-72B · Video-MME K=16 | accuracy: 62.93 -> 66.04 | Improved |
| add query decomposition (w/ Decomp. vs w/o) (ablation) | Top-K selection, Qwen2.5-VL-72B · Video-MME K=64 | accuracy: 68.56 -> 70.03 | Improved |
| use SCOPE instead of AKS | Qwen2.5-VL-72B · Video-MME K=16 | accuracy: 64.2 -> 66.0 | Improved |
| use SCOPE instead of AKS | Qwen2.5-VL-32B · Video-MME K=16 | accuracy: 60.5 -> 63.0 | Improved |
| use SCOPE instead of Uniform sampling | Qwen2.5-VL-72B · Video-MME K=16 | accuracy: 62.6 -> 66.0 | Improved |
| use SCOPE instead of Top-K | Qwen2.5-VL-32B · LongVideoBench K=64 | accuracy: 62.9 -> 64.0 | Improved |
| use SCOPE instead of Top-K (edge latency) | Qwen2.5-VL-72B · Video-MME K=16 | latency (s): 14.26 -> 15.26 | Hurt |
| use edge-cloud SCOPE instead of pure cloud | Qwen2.5-VL-72B · Video-MME K=16 | latency (s): 154.22 -> 23.94 | Improved |
Closest papers by meaning
- VideoReasonBench: Can MLLMs Perform Vision-Centric Complex Video Reasoning? (ICLR 2026 · similarity 0.88)
- SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models (ICML 2026 · similarity 0.88)
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.87)
- LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding (NeurIPS 2024 · similarity 0.87)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.87)
- Logic-in-Frames: Dynamic Keyframe Search via Visual Semantic-Logical Verification for Long Video Understanding (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.
