Research map / Video / Video-language understanding
OASIS: On-Demand Hierarchical Event Memory for Streaming Video Reasoning
Zhijia Liang, Jiaming Li, Weikai Chen, Yanhao Zhang, Haonan Lu, Guanbin Li
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 fine-reasoning retrieval (RAG stage) (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 55.22 -> 57.21 | Improved |
| generate retrieval cue from history instead of naive question retrieval (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 56.13 -> 57.21 | Improved |
| bound root nodes (hierarchical event forest) vs flatten memory (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Perception | accuracy: 76.70 -> 78.14 | Improved |
| bound root nodes (hierarchical event forest) vs flatten memory (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 55.78 -> 57.21 | Improved |
| add event forest to medium buffer and short window (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 54.15 -> 57.21 | Improved |
| add short window to medium buffer and event forest (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Perception | accuracy: 70.37 -> 78.14 | Improved |
| add medium buffer to event forest and short window (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 48.65 -> 57.21 | Improved |
| increase retrieved event nodes k_f from 1 to 2 (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Perception | accuracy: 77.12 -> 78.14 | Improved |
| increase retrieved event nodes k_f from 2 to 3 (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Backward | accuracy: 57.21 -> 56.90 | No clear effect |
| vary merge penalty lambda from 0.0 to 1.0 (ablation) | OASIS (Qwen3-VL-8B) · OVO-Bench Perception | accuracy: 78.26 -> 78.65 | No clear effect |
| increase retrieved QA pairs k_q from 1 to 3 (ablation) | OASIS (Qwen3-VL-8B) · StreamingBench SQA | accuracy: 48.40 -> 49.20 | Improved |
| add OASIS to Qwen3-VL-8B | Qwen3-VL-8B · OVO-Bench Perception | accuracy: 66.79 -> 78.14 | Improved |
Closest papers by meaning
- What Should a Streaming Video Model Remember? (NeurIPS 2026 · similarity 0.89)
- Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models (CVPR 2026 · similarity 0.88)
- CoRDS: Coreset-Based Representative and Diverse Selection for Streaming Video Understanding (NeurIPS 2026 · similarity 0.87)
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.87)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.86)
- 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.
