Research map / Video / Video-language understanding
Test-time Counterfactual Calibration for Hallucination-Resistant Temporal Grounding
Chufan YI, Hongyu Qu, Shiyu Xuan, Rui Yan, Xiangbo Shu, Fang Zhao, Guosen Xie
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 |
|---|---|---|---|
| apply HRVTG test-time counterfactual calibration to the only-positive model | Qwen2.5-VL-3B · Charades-RF | F1_Balanced: 45.12 -> 68.30 | Improved |
| apply HRVTG | Qwen2.5-VL-3B · ActivityNet-RF | F1_Balanced: 60.51 -> 83.91 | Improved |
| apply HRVTG | Qwen2.5-VL-3B · TVGBench-RF | F1_Balanced: 9.78 -> 58.83 | Improved |
| apply HRVTG (consistency anchor reward keeps grounding) | Qwen2.5-VL-3B · Charades-RF | conditional mIoU: 41.48 -> 43.29 | Improved |
| apply HRVTG | Qwen2.5-VL-7B · Charades-RF | conditional mIoU: 48.33 -> 48.94 | No clear effect |
| apply HRVTG | Qwen2.5-VL-7B · ActivityNet-RF | R@0.7: 14.04 -> 13.03 | Mixed |
| use asymmetric KL weight rho=4 instead of symmetric rho=1 (ablation) | HRVTG (3B) · TVGBench-RF | F1_Balanced: 53.16 -> 58.83 | Improved |
| increase rho from 4 to 8 (ablation) | HRVTG (3B) · TVGBench-RF | mIoU / F1_Balanced | Hurt |
| raise TTA learning rate above best value (ablation) | HRVTG (3B) · TVGBench-RF | F1_Balanced vs conditional mIoU: best F1_Balanced 58.83 at 5e-5 | Mixed |
| train only on positive samples (always-answer) | Qwen2.5-VL-3B · TVGBench-RF | F1_Refuse: F1_Rfs 5.28 | Hurt |
Closest papers by meaning
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.89)
- Uncertainty-quantified Rollout Policy Adaptation for Unlabelled Cross-domain Video Temporal Grounding (NeurIPS 2025 · similarity 0.88)
- Ground What You See: Hallucination-Resistant MLLMs via Caption Feedback, Diversity-Aware Sampling, and Conflict Regularization (AAAI 2026 · similarity 0.88)
- Mitigating Hallucination in VideoLLMs via Temporal-Aware Activation Engineering (NeurIPS 2025 · similarity 0.87)
- Grounding or Guessing? Visual Signals for Detecting Hallucinations in Sign Language Translation (ICLR 2026 · similarity 0.87)
- Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling (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.
