Research map / Graph learning / Graph self-supervised learning
Topology-Aware Vision Transformers for Enhanced Scene Recognition
Yunxi Wang, Shuaiyu Liu, Qiling Li, Yazhou Ren, Xiaorong Pu
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 GGM graph embedding via CLS-level concatenation (ablation) | ViT-B · MIT-67 | accuracy (%): 84.18 -> 84.85 | Improved |
| add GAMGN with direct concatenation (ablation) | ViT-B + GGM · MIT-67 | accuracy (%): 84.85 -> 92.99 | Improved |
| add TAG mask-weighted patch embeddings (ablation) | ViT-B + GGM + GAMGN · MIT-67 | accuracy (%): 92.99 -> 98.58 | Improved |
| add all topology modules (GGM, GAMGN, TAG) (ablation) | ViT-B · MIT-67 | accuracy (%): 84.18 -> 98.58 | Improved |
| raise GAMGN learning rate from 1e-1 to 1e-3 (ablation) | TANSR · MIT-67 | accuracy (%): 92.71 -> 98.58 | Improved |
| lower GAMGN learning rate from 1e-3 to 1e-4 (ablation) | TANSR · MIT-67 | accuracy (%): 98.58 -> 97.90 | Hurt |
| raise GAMGN learning rate from 1e-1 to 1e-3 (ablation) | TANSR · Scene-15 | accuracy (%): 96.11 -> 97.12 | Improved |
| increase batch size from 75 to 125 (ablation) | TANSR · MIT-67 | accuracy (%): 98.58 -> 97.48 | Hurt |
| increase batch size from 75 to 150 (ablation) | TANSR · Scene-15 | accuracy (%): 97.12 -> 97.03 | No clear effect |
| enlarge GAT from 16→8 to 32→16 (ablation) | TANSR · MIT-67 | accuracy (%): 97.02 -> 98.58 | Improved |
| enlarge GAT from 32→16 to 64→32 (ablation) | TANSR · MIT-67 | accuracy (%): 98.58 -> 98.46 | No clear effect |
| add TANSR topology modules (compute overhead) | ViT-B · MIT-67 | accuracy (%): 3.88 ± 0.20 -> 8.17 ± 0.35 | Mixed |
Closest papers by meaning
- Linear Transformer Topological Masking with Graph Random Features (ICLR 2025 · similarity 0.86)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.86)
- Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models (NeurIPS 2026 · similarity 0.85)
- Multiview Scene Graph (NeurIPS 2024 · similarity 0.85)
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.85)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 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.
