Research map / Video / Video-language understanding
Learning to Refuse: Refusal-Aware Reinforcement Fine-Tuning for Hard-Irrelevant Queries in Video Temporal Grounding
Jin-Seop Lee, SungJoon Lee, SeongJun Jung, Boyang Li, Jee-Hyong Lee
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 |
|---|---|---|---|
| train GRPO on simply-shuffled queries only (no HI-VTG) (ablation) | Time-R1 · HI-ActivityNet (Time-R1) | F1 avg: 70.5 -> 70.0 | No clear effect |
| add HI-VTG hard-irrelevant data with refuse-IoU reward (ablation) | Time-R1 · HI-ActivityNet (Time-R1) | F1 avg: 70.0 -> 75.1 | Improved |
| add explain reward (ablation) | Time-R1 + GRPO R.IoU · HI-ActivityNet (Time-R1) | F1 avg: 75.1 -> 75.6 | Improved |
| add explain reward (ablation) | Time-R1 + GRPO R.IoU · HI-ActivityNet (Time-R1) | LLM explanation score: 2.29 -> 2.37 | Improved |
| add query correction reward (ablation) | Time-R1 + GRPO R.IoU-Exp · HI-ActivityNet (Time-R1) | F1 avg: 75.6 -> 76.3 | Improved |
| add query correction reward (ablation) | Time-R1 + GRPO R.IoU-Exp · HI-ActivityNet (Time-R1) | RA-IoU mIoU: 52.2 -> 51.9 | No clear effect |
| replace RA-RFT with SFT on HI-VTG (ablation) | Time-R1 · HI-ActivityNet (Time-R1) | RA-IoU mIoU: 51.9 -> 20.9 | Hurt |
| add RA-RFT | Time-R1 · HI-ActivityNet | RA-IoU mIoU: 45.9 -> 51.9 | Improved |
| add RA-RFT | VideoChat-R1 · HI-ActivityNet | RA-IoU mIoU: 39.4 -> 46.5 | Improved |
| add RA-RFT | Time-R1 · HI-TVGBench | RA-IoU mIoU: 37.3 -> 46.6 | Improved |
| add RA-RFT | Qwen2.5-VL-7B · HI-ActivityNet | RA-IoU mIoU: 27.8 -> 49.6 | Improved |
| add RA-RFT | Time-R1 · HI-ActivityNet (strong hard-irrelevant) | F1: 60.0 -> 70.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Time-R1 — Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025) · better on 5 of 5 shared results
- TimeChat — TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding (CVPR 2024) · better on 1 of 1 shared result
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
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · similarity 0.88)
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.88)
- Searching Videos as Trees: Self-Correcting Agents for Grounded Long Video QA (NeurIPS 2026 · similarity 0.87)
- VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding (AAAI 2025 · similarity 0.87)
- EvoGround: Self-Evolving Video Agents for Video Temporal Grounding (NeurIPS 2026 · similarity 0.87)
- TempSamp-R1: Effective Temporal Sampling with Reinforcement Fine-Tuning for Video LLMs (NeurIPS 2025 · similarity 0.87)
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.
