Research map / Video / Video-language understanding
STVG-R1: Incentivizing Instance-Level Reasoning and Grounding in Videos via Reinforcement Learning
Xiaowen Zhang, Zhi Gao, Licheng Jiao, Lingling Li, Qing Li
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 object-centric visual prompts (zero-shot) (ablation) | Qwen3-VL-8B · HCSTVG-v1 test | vIoU@0.3: 28.2 -> 56.5 | Improved |
| add object-centric visual prompts (zero-shot) (ablation) | Qwen2.5-VL-7B · HCSTVG-v1 test | m_tIoU: 40.3 -> 38.7 | Hurt |
| add object-centric visual prompts (zero-shot) (ablation) | Qwen2.5-VL-7B · HCSTVG-v1 test | vIoU@0.3: 28.2 -> 40.7 | Improved |
| add object-centric visual prompts (zero-shot) (ablation) | Qwen2.5-VL-7B · HCSTVG-v2 val | m_vIoU: 19.3 -> 19.5 | No clear effect |
| add GRPO with temporal reward only (no visual prompts) (ablation) | Qwen2.5-VL-7B · HCSTVG-v1 test | m_tIoU: 40.3 -> 57.5 | Improved |
| add GRPO on top of visual prompts instead of SFT (ablation) | Qwen2.5-VL-7B · HCSTVG-v1 test | m_vIoU: 34.3 -> 39.1 | Improved |
| add visual-prompt SFT instead of GRPO (ablation) | Qwen2.5-VL-7B · ST-Align STVG | m_tIoU: 38.3 -> 36.6 | Hurt |
| use coupled spatio-temporal reward instead of sparse spatial reward (ablation) | STVG-R1 · HCSTVG-v1 test | m_vIoU: 39.1 -> 38.3 | Hurt |
| use continuous per-frame IoU spatial reward (ablation) | STVG-R1 · HCSTVG-v1 test | m_vIoU: 39.1 -> 38.6 | Hurt |
| use mixed number+letter prompts instead of numbers (ablation) | zero-shot Qwen2.5-VL-7B · HCSTVG-v1 | m_vIoU: 24.9 -> 15.7 | Hurt |
| increase prompt font size from 20 to 40 (ablation) | zero-shot Qwen2.5-VL-7B · HCSTVG-v1 | vIoU@0.3: 40.6 -> 36.4 | Hurt |
| raise mask filtering threshold to 1/2 from 1/3 (ablation) | data quality and zero-shot · HCSTVG-v1 | m_vIoU (upper bound): 67.7 -> 65.1 | Hurt |
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.
- SpaceVLLM-7B — SpaceVLLM: Endowing Multimodal Large Language Model with Spatio-Temporal Video Grounding Capability (AAAI 2026) · better on 4 of 5 shared results
- VideoGlaMM — VideoGLaMM : A Large Multimodal Model for Pixel-Level Visual Grounding in Videos (CVPR 2025) · 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 0.91)
- Context-Guided Spatio-Temporal Video Grounding (CVPR 2024 · similarity 0.90)
- Unleashing the Potential of Multimodal LLMs for Zero-Shot Spatio-Temporal Video Grounding (NeurIPS 2025 · similarity 0.89)
- Grounded Reinforcement Learning for Visual Reasoning (NeurIPS 2025 · similarity 0.89)
- STPro: Spatial and Temporal Progressive Learning for Weakly Supervised Spatio-Temporal Grounding (CVPR 2025 · similarity 0.89)
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · 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.
