Research map / Graph learning / Knowledge and temporal graphs
Mixture-Kernel Graph Attention Network for Situation Recognition
Mohammed Suhail, Leonid Sigal
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 graph attention to GGNN (ablation) | GGNN · imSitu test | mean score: 38.19 -> 39.89 | Improved |
| add hard-assigned multi-kernel to GGNN (ablation) | GGNN · imSitu test | mean score: 38.19 -> 39.27 | Improved |
| add attention to multi-kernel GGNN (ablation) | GGNN + multi-kernel · imSitu test | mean score: 39.27 -> 40.99 | Improved |
| replace hard multi-kernel with mixture-kernel (convex combination of shared basis) with attention (ablation) | GGNN + attn · imSitu test | mean score: 40.99 -> 42.66 | Improved |
| full model (mixture-kernel + attention) vs naive GGNN, ground truth verb (ablation) | GGNN · imSitu test | value (ground truth verb): 67.32 -> 72.92 | Improved |
| increase propagation steps from T=0 to T=4 (ablation) | Ours · imSitu | value: 58.13 -> 72.93 | Improved |
| increase propagation steps from T=3 to T=4 (ablation) | Ours · imSitu | value: 72.18 -> 72.93 | Improved |
| use predicted basis weights instead of random permuted weights (value) (ablation) | Ours · imSitu (GT verb) | value: 64.26 -> 72.96 | Improved |
| use predicted basis weights instead of random permuted weights (value-all) (ablation) | Ours · imSitu (GT verb) | value-all: 26.85 -> 42.35 | Improved |
| keep all edges rather than removing 75% at random (ablation) | Ours · imSitu (GT verb) | value: 61.18 -> 72.93 | Improved |
| Ours vs Fully-connected Graph with top-1 predicted verb | Ours · imSitu test | value: 27.52 -> 35.41 | Improved |
| Ours vs Fully-connected Graph with top-5 predicted verb | Ours · imSitu test | value: 45.39 -> 55.62 | Improved |
Closest papers by meaning
- Situation Recognition With Graph Neural Networks (ICCV 2017 · similarity 0.91)
- Situation Recognition: Visual Semantic Role Labeling for Image Understanding (CVPR 2016 · similarity 0.88)
- Dependency Structure-Enhanced Graph Attention Networks for Event Detection (AAAI 2024 · similarity 0.86)
- Dynamic Graph Message Passing Networks for Visual Recognition (TPAMI 2022 · similarity 0.86)
- Representation Learning on Visual-Symbolic Graphs for Video Understanding (ECCV 2020 · similarity 0.86)
- Unified Graph Structured Models for Video Understanding (ICCV 2021 · similarity 0.86)
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.
