Research map / Architectures and efficiency / CNN design
LIP: Local Importance-Based Pooling
Ziteng Gao, Limin Wang, Gangshan Wu
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 LIP at the max pooling layer (ablation) | LIP-ResNet-50 (Bottleneck-128) · ImageNet-1k val | top-1 accuracy: 77.87 -> 78.19 | Improved |
| add LIP at Res3 downsampling (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 77.78 -> 77.87 | Improved |
| add LIP at Res4 downsampling (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.92 -> 77.78 | Improved |
| add LIP at Res5 downsampling (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.40 -> 76.92 | Improved |
| add affine instance normalization to logit module (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 77.81 -> 78.19 | Improved |
| add amplified sigmoid to logit module (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 77.89 -> 78.19 | Improved |
| use Bottleneck-64 logit module instead of projection (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 77.49 -> 77.92 | Improved |
| widen logit module from Bottleneck-128 to Bottleneck-256 (ablation) | LIP-ResNet-50 · ImageNet-1k val | top-1 accuracy: 78.19 -> 78.15 | No clear effect |
| replace average pooling with LIP Bottleneck-128 | ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.96 -> 78.19 | Improved |
| replace strided convolution with average pooling | ResNet-50 · ImageNet-1k val | top-1 accuracy: 76.40 -> 76.96 | Improved |
| replace downsampling with LIP | ResNet-101 · ImageNet-1k val | top-1 accuracy: 77.98 -> 79.33 | Improved |
| replace downsampling with LIP | DenseNet-BC-121 · ImageNet-1k val | top-1 accuracy: 75.62 -> 76.64 | Improved |
Closest papers by meaning
- Detail-Preserving Pooling in Deep Networks (CVPR 2018 · similarity 0.88)
- S3Pool: Pooling With Stochastic Spatial Sampling (CVPR 2017 · similarity 0.88)
- Gaussian-Based Pooling for Convolutional Neural Networks (NeurIPS 2019 · similarity 0.88)
- Global Feature Guided Local Pooling (ICCV 2019 · similarity 0.87)
- Wavelet Pooling for Convolutional Neural Networks (ICLR 2018 · similarity 0.86)
- Building Detail-Sensitive Semantic Segmentation Networks With Polynomial Pooling (CVPR 2019 · similarity 0.86)
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.
