Research map / Video / Video-language understanding
CurveStream: Boosting Streaming Video Understanding in MLLMs via Curvature-Aware Hierarchical Visual Memory Management
Chao Wang, Xudong Tan, Jianjian Cao, Kangcong Li, Tao Chen
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 curvature score (CAS) to the memory (ablation) | Qwen2-VL-7B · StreamingBench | accuracy: 69.04 -> 78.16 | Improved |
| add hierarchical memory (HVMM) without CAS (ablation) | Qwen2-VL-7B · StreamingBench | accuracy: 69.04 -> 78.80 | Improved |
| add both CAS and HVMM (ablation) | Qwen2-VL-7B · StreamingBench | accuracy: 78.16 -> 81.04 | Improved |
| add CAS to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · OVOBench | accuracy: 70.10 -> 78.49 | Improved |
| add HVMM to Qwen3-VL-8B (ablation) | Qwen3-VL-8B · OVOBench | accuracy: 70.10 -> 74.79 | Improved |
| add both CAS and HVMM (ablation) | Qwen3-VL-8B · OVOBench | accuracy: 78.49 -> 80.76 | Improved |
| use curvature instead of cosine similarity for frame selection (ablation) | Qwen2-VL-7B · StreamingBench (N=10) | accuracy: 73.28 -> 77.31 | Improved |
| use curvature instead of dense optical flow (ablation) | Qwen2-VL-7B · StreamingBench (N=10) | accuracy: 46.54 -> 77.31 | Improved |
| use curvature instead of pyramid optical flow (ablation) | Qwen2-VL-7B · StreamingBench (N=10) | accuracy: 75.69 -> 77.31 | Improved |
| use curvature instead of uniform sampling (ablation) | Qwen2-VL-7B · StreamingBench (N=10) | accuracy: 69.04 -> 77.31 | Improved |
| vary curvature weight lambda from 0.4 to 0.2 (ablation) | Qwen2-VL-7B · OVOBench | accuracy: 62.50 -> 65.83 | Mixed |
| integrate CurveStream on Qwen2.5-VL-7B | Qwen2.5-VL-7B · VideoMME | accuracy: 64.52 -> 62.97 | Hurt |
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.
- + Freshmem — FreshMem: Brain-Inspired Frequency-Space Hybrid Memory for Streaming Video Understanding (NeurIPS 2026) · better on 2 of 2 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
- StreamForest: Efficient Online Video Understanding with Persistent Event Memory (NeurIPS 2025 · similarity 0.90)
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.89)
- V-CAST: Video Curvature-Aware Spatio-Temporal Pruning for Efficient Video Large Language Models (NeurIPS 2026 · similarity 0.88)
- FreshMem: Brain-Inspired Frequency-Space Hybrid Memory for Streaming Video Understanding (NeurIPS 2026 · similarity 0.88)
- Streaming Long Video Understanding with Large Language Models (NeurIPS 2024 · similarity 0.88)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.88)
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.
