Research map / Video / Action recognition
Invert4TVG: A Temporal Video Grounding Framework with Inversion Tasks Preserving Action Understanding Ability
Chenzhaoyu, Hongnan Lin, Yongwei Nie, Fei Ma, Xuemiao Xu, Fei Yu, 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 Invert-TVG auxiliary tasks to Time-R1* training (3B) | Qwen2.5-VL-3B RL · Charades-STA | R1@0.7: 36.9 -> 44.0 | Improved |
| add Invert-TVG auxiliary tasks (7B) | Qwen2.5-VL-7B RL · Charades-STA | R1@0.7: 50.1 -> 51.4 | Improved |
| add all three inversion tasks versus only video description (ablation) | Invert4TVG · Charades-STA | R1@0.7: 39.4 -> 44.0 | Improved |
| add all three inversion tasks versus only action recognition (ablation) | Invert4TVG · Charades-STA | R1@0.7: 43.8 -> 44.0 | Improved |
| add all three inversion tasks versus only verb completion (ablation) | Invert4TVG · Charades-STA | R1@0.5: 68.0 -> 69.0 | Improved |
| add video description to VC+AR (ablation) | Invert4TVG · Charades-STA | R1@0.7: 43.8 -> 44.0 | Improved |
| add action recognition to VC+VD (ablation) | Invert4TVG · Charades-STA | R1@0.7: 42.1 -> 44.0 | Improved |
| add verb completion to AR+VD (ablation) | Invert4TVG · Charades-STA | R1@0.7: 43.6 -> 44.0 | Improved |
| add only inversion task VD to TVG-only training (ablation) | Time-R1* 3B · Charades-STA | R1@0.7: 36.9 -> 39.4 | Improved |
| use binary 0/1 Invert-TVG reward instead of cosine similarity (ablation) | Invert4TVG · Charades-STA | R1@0.7: 39.8 -> 44.0 | Improved |
| use binary reward instead of cosine similarity (ablation) | Invert4TVG · Charades-STA | R1@0.5: 62.2 -> 69.0 | Improved |
| add Invert-TVG tasks within same training time budget | Invert4TVG · Charades-STA | training time: 80 hours -> 80 hours | 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.
- VideoChat-Flash — VideoChat-Flash: Hierarchical Compression for Long-Context Video Modeling (ICLR 2026) · 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
- Time-R1: Post-Training Large Vision Language Model for Temporal Video Grounding (NeurIPS 2025 · similarity 0.90)
- STVG-R1: Incentivizing Instance-Level Reasoning and Grounding in Videos via Reinforcement Learning (ICLR 2026 · similarity 0.90)
- VideoGEM: Training-free Action Grounding in Videos (CVPR 2025 · similarity 0.90)
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · similarity 0.89)
- SARL-STG: A Spatially Aware Reinforcement Learning Framework for Refining MLLMs in Spatio-Temporal Video Grounding (CVPR 2026 · similarity 0.89)
- Learning to Refuse: Refusal-Aware Reinforcement Fine-Tuning for Hard-Irrelevant Queries in Video Temporal Grounding (CVPR 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.
