Research map / Video / Video-language understanding
Open-ended Hierarchical Streaming Video Understanding with Vision Language Models
Hyolim Kang, Yunsu Park, Youngbeom Yoo, Yeeun Choi, Seon Joo Kim
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 progress-drop based action end detection (hybrid boundary detection) (ablation) | OpenHOUSE streaming module · EgoGS step (GT labels) | F1 (loc.): 16.78 -> 39.35 | Improved |
| add hybrid boundary detection (ablation) | OpenHOUSE streaming module · EgEx substep (GT labels) | F1 (loc.): 8.42 -> 52.2 | Improved |
| add hybrid boundary detection (ablation) | OpenHOUSE streaming module · EK100 substep (GT labels) | F1 (loc.): 31.46 -> 48.95 | Improved |
| train on LLM pseudo step labels instead of ground-truth (ablation) | OpenHOUSE streaming module · EgoGS step | F1 (loc.) @tIoU 0.3: 51.58 -> 50.65 | No clear effect |
| sample frames by hierarchy and add previous predictions to context memory (ablation) | OpenHOUSE VLM captioning · EgoGS step captions | F1 (loc.+desc.): 13.08 -> 13.05 with 22770 instead of 59385 frames | No clear effect |
| add previous text predictions as context given hierarchical frames (ablation) | OpenHOUSE VLM captioning · EgoGS step captions | GPT Score: 2.42 -> 2.76 | Improved |
| use lightweight streaming module as VLM caller rather than a VLM on every frame | OpenHOUSE · 46-minute video inference | inference speed: 24 fps, 16 times faster | Improved |
| merge EgoGS and EgEx training data (ablation) | OpenHOUSE · EK100 validation (cross-dataset) | F1 (loc.) @tIoU 0.3, step: 42.49 -> 51.64 (vs EgoGS only) | Improved |
| use OAD-based streaming instead of SDVC fixed decoding points | Hierarchical streaming understanding · EgoGS step | AEDT (s): 36.36 -> 4.94 | Improved |
| use frozen powerful VLM with SDVC proposals instead of fine-tuning SDVC (ablation) | SDVC · EgoGS step | F1 (loc.+desc.) @tIoU 0.3: 2.61 -> 6.20 | Improved |
| use a smaller VLM (InternVL2-8B instead of InternVL2-40B-AWQ) (ablation) | OpenHOUSE captioning · EgoGS step | F1 (loc.+desc.) @tIoU 0.3: 15.23 -> 10.01 | Hurt |
| use GPT-4o instead of InternVL2-40B-AWQ (ablation) | OpenHOUSE captioning · EgoGS step | F1 (loc.+desc.) @tIoU 0.3: 15.23 -> 15.53 | No clear effect |
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.
- MiniROAD — MiniROAD: Minimal RNN Framework for Online Action Detection (ICCV 2023) · 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 0.88)
- StreamOV: Streaming Omni-Video Understanding via Evidence-Guided Memory and Response Triggering (NeurIPS 2026 · similarity 0.86)
- LiveStar: Live Streaming Assistant for Real-World Online Video Understanding (NeurIPS 2025 · similarity 0.86)
- OmniViD: A Generative Framework for Universal Video Understanding (CVPR 2024 · similarity 0.86)
- Does Video-Text Pretraining Help Open-Vocabulary Online Action Detection? (NeurIPS 2024 · similarity 0.86)
- HyrCap: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding (NeurIPS 2026 · 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.
