Research map / Video / Video-language understanding
HiTeA: Hierarchical Temporal Alignment for Training-Free Long-Video Temporal Grounding
Xinyi Xu, Hongsong Wang, Guo-Sen Xie, Caifeng Shan, Fang Zhao
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 Candidate Refinement (CR) to three-level HTD (ablation) | HiTeA · TACoS | mIoU: 16.99 -> 19.79 | Improved |
| add Candidate Refinement (CR) to three-level HTD (ablation) | HiTeA · Charades-STA | mIoU: 41.61 -> 46.29 | Improved |
| add Hierarchical Temporal Decomposition (event level) over uniform segmentation, no CR (ablation) | HiTeA · Charades-STA | mIoU: 35.07 -> 41.28 | Improved |
| extend decomposition from one level to three levels (event+scene+action), no CR (ablation) | HiTeA · TACoS | mIoU: 15.64 -> 16.99 | Improved |
| extend decomposition to three levels, no CR (ablation) | HiTeA · Charades-STA | mIoU: 41.28 -> 41.61 | No clear effect |
| use scene-level only decomposition instead of event-level (ablation) | HiTeA · TACoS | mIoU: 15.64 -> 13.01 | Hurt |
| add CR to uniform segmentation without HTD (ablation) | HiTeA · Charades-STA | mIoU: 35.07 -> 43.93 | Improved |
| use Qwen2.5-VL 7B instead of 3B as scorer (ablation) | HiTeA · TACoS | mIoU: 16.36 -> 19.79 | Improved |
| add a large VLM scorer to VideoCLIP-XL alone (ablation) | HiTeA · TACoS | mIoU: 9.50 -> 19.79 | Improved |
| swap Qwen2.5-VL 7B for InternVideo2.5 7B (ablation) | HiTeA · Charades-STA | mIoU: 46.29 -> 45.94 | No clear effect |
| enable Hierarchical Merging on short videos (ablation) | HiTeA · Charades-STA | mIoU: 46.29 -> 45.09 | Hurt |
| add VideoCLIP-based pre-filtering to reduce candidate segments | HiTeA · Ego4D-NLQ | segments per video: 146.68 -> 9.0 | Improved |
Closest papers by meaning
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.90)
- ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos (CVPR 2025 · similarity 0.89)
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · similarity 0.89)
- Empower Words: DualGround for Structured Phrase and Sentence-Level Temporal Grounding (NeurIPS 2025 · similarity 0.89)
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.89)
- TOPA: Extending Large Language Models for Video Understanding via Text-Only Pre-Alignment (NeurIPS 2024 · similarity 0.89)
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.
