Research map / Video / Video-language understanding
Asymmetric Relation Consistency Reasoning for Video Relation Grounding
Huan Li, Ping Wei, Jiapeng Li, Zeyu Ma, Jiahui Shang, Nanning Zheng
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 asymmetric (conditional) subject-object reasoning (ablation) | ARC · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 24.86 -> 26.42 | Improved |
| add asymmetric reasoning (ablation) | ARC · ImageNet-VidVRD (sIoU=0.7) | AccR: 16.13 -> 20.44 | Improved |
| add relation consistency KL loss (ablation) | ARC · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 23.61 -> 26.42 | Improved |
| add relation consistency KL loss (ablation) | ARC · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccS: 35.99 -> 39.66 | Improved |
| add relation consistency KL loss (ablation) | ARC · ImageNet-VidVRD (sIoU=0.7) | AccO: 31.13 -> 33.39 | Improved |
| add relation-aware temporal reasoning (RTR) (ablation) | ARC · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 23.86 -> 26.42 | Improved |
| add RTR (helps average and strict thresholds, slightly lower at sIoU=0.3) (ablation) | ARC · ImageNet-VidVRD (sIoU=0.3) | AccR: 33.03 -> 32.53 | Mixed |
| use IoU linking cost in inference | ARC · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 23.75 -> 26.42 | Improved |
| use IoU linking cost in inference | vRGV · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 20.54 -> 24.58 | Improved |
| replace transformer baseline with ARC | Our baseline · ImageNet-VidVRD (average over sIoU 0.3/0.5/0.7) | AccR: 24.38 -> 26.42 | Improved |
| ARC vs vRGV in zero-shot evaluation | ARC · ImageNet-VidVRD (zero-shot) | AccR: 10.27 -> 11.19 | Improved |
| ARC vs vRGV in zero-shot subject accuracy | ARC · ImageNet-VidVRD (zero-shot) | AccS: 18.94 -> 17.34 | Hurt |
Closest papers by meaning
- Visual Relation Grounding in Videos (ECCV 2020 · similarity 0.88)
- Inverse Compositional Learning for Weakly-supervised Relation Grounding (ICCV 2023 · similarity 0.87)
- Multi-Pair Temporal Sentence Grounding via Multi-Thread Knowledge Transfer Network (AAAI 2025 · similarity 0.85)
- VideoMind: A Chain-of-LoRA Agent for Temporal-Grounded Video Reasoning (ICLR 2026 · similarity 0.85)
- Where Does It Exist: Spatio-Temporal Video Grounding for Multi-Form Sentences (CVPR 2020 · similarity 0.85)
- Cross-modal Causal Relation Alignment for Video Question Grounding (CVPR 2025 · similarity 0.85)
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.
