Research map / Graph learning / Knowledge and temporal graphs
Unifying Nonlocal Blocks for Neural Networks
Lei Zhu, Qi She, Duo Li, Yanye Lu, Xuejing Kang, Jie Hu, Changhu Wang
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 symmetrized, normalized affinity matrix to the NL block (SNL_a1) (ablation) | NL block in ResNet50-FCN · Cityscapes val | mIoU (%): 74.15 -> 75.76 | Improved |
| add the complete 1st-order Chebyshev term (separate weights for identity and affinity terms) (SNL_a2) (ablation) | NL block in ResNet50-FCN · Cityscapes val | mIoU (%): 74.15 -> 75.94 | Improved |
| combine symmetric affinity with complete 1st-order Chebyshev term (full SNL) (ablation) | SNL_a1 / SNL_a2 · Cityscapes val | mIoU (%): 75.76 -> 76.87 | Improved |
| replace NL with SNL block | PreResNet56 (stage 1) · CIFAR-100 | Top-1 accuracy (%): 75.31 -> 76.41 | Improved |
| increase transferred channels to non-reduction (Cs=C1) for CGNL (ablation) | CGNL block · CIFAR-100 | Top-1 accuracy (%): 75.75 -> 74.71 | Hurt |
| increase transferred channels to non-reduction (Cs=C1) for SNL (ablation) | SNL block · CIFAR-100 | Top-1 accuracy (%): 76.41 -> 76.34 | No clear effect |
| move SNL block from stage 1 to stage 3 (ablation) | PreResNet56 + SNL · CIFAR-100 | Top-1 accuracy (%): 76.41 -> 75.68 | Hurt |
| use three SNL blocks (one per stage) instead of one (ablation) | PreResNet56 + SNL · CIFAR-100 | Top-1 accuracy (%): 76.41 -> 76.70 | Improved |
| use three NL blocks instead of one (ablation) | PreResNet56 + NL · CIFAR-100 | Top-1 accuracy (%): 75.31 -> 74.34 | Hurt |
| add SNL block to ResNet50 | ResNet50 · ImageNet | Top-1 accuracy (%): 76.15 -> 78.11 | Improved |
| add NL block to ResNet50 | ResNet50 · CIFAR-10 | Top-1 accuracy (%): 94.94 -> 94.01 | Hurt |
| add SNL block to I3D | I3D · UCF-101 | Top-1 accuracy (%): 81.57 -> 84.39 | Improved |
Closest papers by meaning
- Nonlocal Neural Networks, Nonlocal Diffusion and Nonlocal Modeling (NeurIPS 2018 · similarity 0.88)
- Compact Generalized Non-local Network (NeurIPS 2018 · similarity 0.85)
- Disentangled Non-local Neural Networks (ECCV 2020 · similarity 0.85)
- Position: Spectral GNNs Rely Less on Graph Fourier Basis than Conceived (ICML 2025 · similarity 0.82)
- A General Graph Spectral Wavelet Convolution via Chebyshev Order Decomposition (ICML 2025 · similarity 0.82)
- HoloNets: Spectral Convolutions do extend to Directed Graphs (ICLR 2024 · similarity 0.82)
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.
