Research map / Video / Video segmentation and tracking
Compact Generalized Non-local Network
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 CGNL blocks instead of the same number of non-local blocks | ResNet-50 and ResNet-101 · Mini-Kinetics | top1 accuracy: 76.53 -> 77.76; 77.53 -> 78.79; 78.02 -> 79.54; 79.21 -> 79.88 | Helped |
| Add CGNL blocks instead of the same number of non-local blocks | ResNet-50 and ResNet-101 · UCF101 | top1 accuracy: 84.79 -> 85.14; 85.10 -> 85.68; 85.49 -> 86.35; 86.10 -> 86.24 | Helped |
| Add 1 CGNL block to Mask R-CNN | Mask R-CNN with ResNet-50 backbone · COCO2017 | box AP: 35.02 -> 35.70 | Helped |
| Add 1 dot-product CGNL block | ResNet-50 and ResNet-152 · ImageNet validation | top1 accuracy: 76.15 -> 77.69; 78.31 -> 79.53 | Helped |
| Use Gaussian RBF instead of dot-product kernel (ablation) | CGNL · CUB-200-2011 | top1 accuracy: 85.14 -> 84.10 | Hurt |
| Use embedded Gaussian instead of dot-product kernel (ablation) | CGNL · CUB-200-2011 | top1 accuracy: 85.14 -> 84.01 | Hurt |
| Add channel grouping to the 1-block CGNL configuration (ablation) | ResNet-101 + 1 CGNL · Mini-Kinetics | top1 accuracy: 86.17; 86.24; 86.35; 86.13; 86.04 | Mixed |
| Increase grouping from 8 to 16 groups (ablation) | ResNet-101 + 5 CGNL · Mini-Kinetics | top1 accuracy: 86.24 -> 86.43 | Helped |
| Add a simple residual block instead of CGNL (ablation) | ResNet-50 · CUB-200-2011 | top1 accuracy: 85.14 vs 84.11 | Hurt |
Closest papers by meaning
- Non-Local Neural Networks (CVPR 2018 · similarity 0.90)
- LatentGNN: Learning Efficient Non-local Relations for Visual Recognition (ICML 2019 · similarity 0.85)
- Memory Enhanced Global-Local Aggregation for Video Object Detection (CVPR 2020 · similarity 0.84)
- Few-Shot Learning With Localization in Realistic Settings (CVPR 2019 · similarity 0.84)
- StarNet: towards Weakly Supervised Few-Shot Object Detection (AAAI 2021 · similarity 0.84)
- Accelerating Non-Maximum Suppression: A Graph Theory Perspective (NeurIPS 2024 · similarity 0.83)
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.