Research map / Video / Video-language understanding
TimeExpert: An Expert-Guided Video LLM for Video Temporal Grounding
Zuhao Yang, Yingchen Yu, Yunqing Zhao, Shijian Lu, Song Bai
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 token-adaptive routing (ablation) | TimeExpert · YouCook2 (zero-shot) | CIDEr: 7.6 -> 8.2 | Improved |
| add token-adaptive routing (ablation) | TimeExpert · Charades-STA (zero-shot) | R@1 IoU=0.5: 40.5 -> 42.8 | Improved |
| add token-adaptive routing (ablation) | TimeExpert · QVHighlights (zero-shot) | HIT@1: 42.6 -> 46.9 | Improved |
| add task-dependent auxiliary loss (ablation) | TimeExpert · QVHighlights (zero-shot) | HIT@1: 45.2 -> 46.9 | Improved |
| add task-dependent auxiliary loss (ablation) | TimeExpert · Charades-STA (zero-shot) | R@1 IoU=0.5: 41.3 -> 42.8 | Improved |
| add separate time/score encoders and decoding heads (ablation) | TimeExpert · VTG tasks | instruction following | Improved |
| increase input frames from 8 to 128 (ablation) | TimeExpert · Charades-STA (zero-shot) | R@1 IoU=0.5: 30.2 -> 41.5 | Improved |
| increase input frames from 64 to 128 (ablation) | TimeExpert · QVHighlights (zero-shot) | HIT@1: 44.9 -> 46.5 | Improved |
| increase activated experts from k=2 to k=6 (ablation) | Vanilla MoE · QVHighlights (zero-shot) | mAP: 29.2 -> 29.6 | Improved |
| increase activated experts from k=6 to k=8 (ablation) | Vanilla MoE · YouCook2 (zero-shot) | F1 score: 23.5 -> 23.4 | No clear effect |
| replace fixed top-2 routing with adaptive-k routing (ablation) | Vanilla MoE · QVHighlights (zero-shot) | HIT@1: 45.8 -> 46.9 | Improved |
| train on TRACE's data recipe instead of full data | TimeExpert · YouCook2 (zero-shot) | CIDEr: 8.2 -> 8.1 | No clear effect |
Closest papers by meaning
- VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding (AAAI 2025 · similarity 0.94)
- TRACE: Temporal Grounding Video LLM via Causal Event Modeling (ICLR 2025 · similarity 0.91)
- Empower Words: DualGround for Structured Phrase and Sentence-Level Temporal Grounding (NeurIPS 2025 · similarity 0.90)
- ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos (CVPR 2025 · similarity 0.90)
- DeCafNet: Delegate and Conquer for Efficient Temporal Grounding in Long Videos (CVPR 2025 · similarity 0.90)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · 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.
