Research map / Graph learning / Knowledge and temporal graphs
Multiscale Graph Texture Network
Ravishankar Evani, Deepu Rajan, Shangbo Mao
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 |
|---|---|---|---|
| use GTN instead of RADAM on frozen ConvNeXt backbones | RADAM · KTH-TIPS2-b | accuracy: 89.3 -> 92.5 (ConvNeXt-L) | Improved |
| use GTN instead of RADAM on GTOS-M | RADAM · GTOS-M | accuracy: 85.8 -> 93.9 (ConvNeXt-L) | Improved |
| use GTN instead of RADAM on FMD | RADAM · FMD | accuracy: 89.3 -> 88.1 (ConvNeXt-L) | Hurt |
| use GTN instead of MPAP with ResNet50 on DTD | MPAP · DTD | accuracy: 78.0 -> 74.6 | Hurt |
| scale up the frozen ConvNeXt backbone size | GTN versus RADAM · DTD | accuracy: 75.4 -> 79.3 (ConvNeXt-N to L) | Improved |
| add residual message passing on top of vertex representation (ablation) | GTN · DTD | accuracy: 77.0 -> 77.5 | Improved |
| add vertex representation of latent texture attributes (ablation) | GTN · DTD | accuracy: 27.6 -> 77.0 | Improved |
| add self-loop re-calibration at r=16 (ablation) | GTN · DTD | accuracy: 71.4 -> 71.5 | No clear effect |
| add more backbone layers as graph sources (M from 1 to 5) (ablation) | GTN · DTD | accuracy: 75.0 -> 77.5 | Improved |
| increase compression ratio r from 2 to 16 (ablation) | GTN · DTD | accuracy: 77.9 -> 71.5 | Hurt |
Closest papers by meaning
- Deep TEN: Texture Encoding Network (CVPR 2017 · similarity 0.87)
- Deep Structure-Revealed Network for Texture Recognition (CVPR 2020 · similarity 0.87)
- Deep Multiple-Attribute-Perceived Network for Real-World Texture Recognition (ICCV 2019 · similarity 0.87)
- MGNNI: Multiscale Graph Neural Networks with Implicit Layers (NeurIPS 2022 · similarity 0.86)
- Chebyshev Attention Depth Permutation Texture Network with Latent Texture Attribute Loss (CVPR 2025 · similarity 0.85)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · 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.
