Research map / Architectures and efficiency / Attention and transformers
Convolutional Neural Networks Driven by Content Similarity
Ligeng Zou, Guihu Zhao
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 sorted 1D depth-wise convolution branch to gated convolution (ablation) | Ego-T · ImageNet-1K val | top-1 accuracy (%): 83.5 -> 84.0 | Improved |
| enlarge 1D window from 6 ln(N)+1 to 24 ln(N)+1 (ablation) | Ego-T · ImageNet-1K val | top-1 accuracy (%): 83.7 -> 84.0 | Improved |
| use global 1D convolution instead of window 24 ln(N)+1 (ablation) | Ego-T · ImageNet-1K val | top-1 accuracy (%): 84.0 -> 84.0 | No clear effect |
| use linspace instead of logspace for weight decay (ablation) | Ego-T · ImageNet-1K val | top-1 accuracy (%): 84.0 -> 83.9 | No clear effect |
| channel-wise sorting of tokens by feature similarity enlarges effective receptive field (ablation) | Ego-T · effective receptive field visualization | influence region (qualitative) | Improved |
| reduce COCO batch size from 16 to 8 | Ego-T · COCO val2017 | APb | Hurt |
| Ego-T vs MogaNet-S on detection | Ego-T · COCO val2017 (Cascade Mask R-CNN) | APb: 51.6 -> 50.7 | Hurt |
| Ego-B vs MogaNet-B on detection | Ego-B · COCO val2017 (Cascade Mask R-CNN) | APb: 52.6 -> 53.3 | Improved |
Closest papers by meaning
- Teaching Compositionality to CNNs (CVPR 2017 · similarity 0.86)
- Container: Context Aggregation Networks (NeurIPS 2021 · similarity 0.86)
- Neural Similarity Learning (NeurIPS 2019 · similarity 0.85)
- Similarity and Matching of Neural Network Representations (NeurIPS 2021 · similarity 0.85)
- Linking in Style: Understanding learned features in deep learning models (ECCV 2024 · similarity 0.85)
- Going Beyond Neural Network Feature Similarity: The Network Feature Complexity and Its Interpretation Using Category Theory (ICLR 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.
