Research map / Video / Video-language understanding
TAR: Temporal Anchor-Constrained Reasoning for Video Temporal Grounding
Chaohong Guo, Xun Mo, Yongwei Nie, Fei Ma, Xuemiao Xu, Chengjiang Long
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 T-anchor reward and anchor reasoning on top of Time-R1 (ablation) | Qwen2.5-VL-7B · Charades-STA (fine-tuned) | R1@0.7: 47.5 -> 50.2 | Improved |
| replace standard IoU reward with soft IoU in Time-R1 (ablation) | Time-R1 7B · Charades-STA (fine-tuned) | R1@0.7: 47.5 -> 48.1 | Improved |
| add SFT cold start on bootstrapped CoT data before GRPO (ablation) | TAR 7B · Charades-STA (fine-tuned) | R1@0.7: 45.9 -> 50.2 | Improved |
| add GRPO after SFT (versus SFT alone) (ablation) | TAR 7B · Charades-STA (fine-tuned) | R1@0.7: 21.0 -> 50.2 | Improved |
| add progressive refinement reward (ablation) | TAR reward · Charades-STA (fine-tuned) | mIoU: 58.5 -> 61.1 | Improved |
| add length penalty TAR_num on anchor count (ablation) | TAR reward · Charades-STA (fine-tuned) | mIoU: 41.1 -> 61.1 | Improved |
| add progressive refinement to sIoU-only reward (ablation) | TAR reward · Charades-STA (fine-tuned) | mIoU: 39.1 -> 41.1 | Improved |
| TAR vs Time-R1 on ActivityNet-Captions zero-shot R1@0.7 | Qwen2.5-VL-7B · ActivityNet-Captions (zero-shot) | R1@0.7: 21.4 -> 19.8 | Hurt |
| TAR vs Time-R1 on TVGBench zero-shot | Qwen2.5-VL-7B · TVGBench (zero-shot) | mIoU: 29.2 -> 30.6 | Mixed |
| TAR reasoning pattern transferred to zero-shot VQA | Qwen2.5-VL-7B · VideoMME (zero-shot VQA) | accuracy: 53.0 -> 54.9 | Improved |
| TAR vs Time-R1 on QVHighlights | Qwen2.5-VL-7B · QVHighlights (fine-tuned) | mIoU: 59.1 -> 65.9 | Improved |
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.
- Time-R1 — Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025) · better on 6 of 9 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
- Temporal-Aware Reasoning Optimization for Video Temporal Grounding (ICML 2026 · similarity 0.93)
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · similarity 0.90)
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.90)
- Motion-Grounded Video Reasoning: Understanding and Perceiving Motion at Pixel Level (CVPR 2025 · similarity 0.89)
- Universal Video Temporal Grounding with Generative Multi-modal Large Language Models (NeurIPS 2025 · similarity 0.89)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 2026 · similarity 0.88)
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.
