Research map / Graph learning / Knowledge and temporal graphs
Graph Representation for Order-Aware Visual Transformation
Yue Qiu, Yanjun Sun, Fumiya Matsuzawa, Kenji Iwata, Hirokatsu Kataoka
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 |
|---|---|---|---|
| Represent change contents as graph nodes and temporal orders as directed edges, explicitly associating changes | VTGen · OVT test | per-scene accuracy: 42.1 -> 57.8 | Improved |
| Represent change contents as graph nodes and temporal orders as directed edges, explicitly associating changes | VTGen · CLEVR-Multi-Change | SPICE: 71.7 -> 81.3 | Improved |
| Use graph matching loss instead of cross-entropy loss (ablation) | VTGen transformer decoder, 1 layer and 4 heads · OVT test | per-change accuracy (with order): 49.6 -> 52.4 | Improved |
| Use a transformer decoder instead of a GCN decoder (ablation) | VTGen with graph matching loss and 1 decoder layer · OVT test | per-change accuracy (with order): 49.0 -> 50.8 | Improved |
| Use one decoder layer instead of two (ablation) | VTGen transformer with graph matching loss and 2 heads · OVT test | per-change accuracy (with order): 36.6 -> 50.8 | Improved |
| Use one decoder layer instead of two (ablation) | VTGen transformer with graph matching loss and 4 heads · OVT test | per-change accuracy (with order): 41.1 -> 52.4 | Improved |
| Use four attention heads instead of two (ablation) | VTGen transformer with graph matching loss and 1 layer · OVT test | per-change accuracy (with order): 50.8 -> 52.4 | Improved |
| Use four attention heads instead of two (ablation) | VTGen transformer with cross-entropy loss and 1 layer · OVT test | per-change accuracy (with order): 50.2 -> 49.6 | Hurt |
| Use graph matching loss instead of cross-entropy loss (ablation) | VTGen transformer decoder, 1 layer and 2 heads · OVT test | per-change accuracy (with order): 50.2 -> 50.8 | Improved |
| Use a two-layer GCN instead of a one-layer GCN (ablation) | VTGen with cross-entropy loss · OVT test | per-change accuracy (with order): 47.2 -> 42.6 | Hurt |
| Use a two-layer GCN instead of a one-layer GCN (ablation) | VTGen with graph matching loss · OVT test | per-change accuracy (with order): 49.0 -> 45.2 | Hurt |
Closest papers by meaning
- Learning to Evolve Scenes: Reasoning about Human Activities with Scene Graphs (NeurIPS 2026 · similarity 0.86)
- A Category Agnostic Model for Visual Rearrangment (CVPR 2024 · similarity 0.86)
- Tracking and Understanding Object Transformations (NeurIPS 2025 · similarity 0.85)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 2024 · similarity 0.84)
- HIG: Hierarchical Interlacement Graph Approach to Scene Graph Generation in Video Understanding (CVPR 2024 · similarity 0.84)
- REOrdering Patches Improves Vision Models (NeurIPS 2025 · similarity 0.84)
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.
