Research map / Video / Video-language understanding
Inverse Compositional Learning for Weakly-supervised Relation Grounding
Huan Li, Ping Wei, Zeyu Ma, 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 holistic inverse learning and partial-level learning to the reconstruction baseline (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 24.10 -> 26.05 | Improved |
| add predicate term of partial-level classification loss (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 22.68 -> 26.05 | Improved |
| add subject term of partial-level classification loss (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 25.06 -> 26.05 | Improved |
| add object term of partial-level classification loss (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 25.07 -> 26.05 | Improved |
| add L+- inverse learning term (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 24.04 -> 26.05 | Improved |
| add L-+ inverse learning term (ablation) | ICL · ImageNet-VidVRD (sIoU=0.3) | AccR (%): 32.20 -> 31.85 at sIoU=0.3; 17.76 -> 18.46 at sIoU=0.7 | Mixed |
| use three-term inverse loss with partly relevant features instead of delta(d++ - d--) only (ablation) | ICL · ImageNet-VidVRD (average) | AccR (%): 24.49 -> 26.05 | Improved |
| add predicate term of partial loss (ablation) | ICL · HICO-DET (sIoU=0.5) | AccR (%): 36.91 -> 46.09 | Improved |
| add object term of partial loss (ablation) | ICL · HICO-DET (sIoU=0.5) | AccR (%): 24.21 -> 46.09 | Improved |
| use inverse loss with three terms instead of single delta term (ablation) | ICL · HICO-DET (sIoU=0.5) | AccR (%): 22.54 -> 46.09 | Improved |
| ICL versus ARC without Viterbi decoding | ICL · ImageNet-VidVRD (average) | AccR (%): 23.75 -> 26.05 | Improved |
| ICL versus ARC with Viterbi decoding | ICL# · ImageNet-VidVRD (average) | AccS and AccR (%): AccS 39.66 -> 39.57; AccR 26.42 -> 27.87 | Mixed |
Closest papers by meaning
- Variational Cross-Graph Reasoning and Adaptive Structured Semantics Learning for Compositional Temporal Grounding (TPAMI 2023 · similarity 0.87)
- Compositional Temporal Grounding With Structured Variational Cross-Graph Correspondence Learning (CVPR 2022 · similarity 0.87)
- Relation-aware Instance Refinement for Weakly Supervised Visual Grounding (CVPR 2021 · similarity 0.87)
- Cross-modal Causal Relation Alignment for Video Question Grounding (CVPR 2025 · similarity 0.86)
- Investigating Compositional Challenges in Vision-Language Models for Visual Grounding (CVPR 2024 · similarity 0.86)
- Local-Global Multi-Modal Distillation for Weakly-Supervised Temporal Video Grounding (AAAI 2024 · 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.
