Research map / Architectures and efficiency / CNN design
Weighted Channel Dropout for Regularization of Deep Convolutional Neural Network
Saihui Hou, Zilei 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 WCD to VGGNet-16 | VGGNet-16 · CUB-200-2011 | top-1 accuracy: 72.31% -> 76.10% | Improved |
| add WCD to ResNet-101 | ResNet-101 · CUB-200-2011 | top-1 accuracy: 76.45% -> 77.22% | Improved |
| add WCD to Inception-V3 | Inception-V3 · Stanford Cars | top-1 accuracy: 93.17% -> 93.41% | Improved |
| add WCD to Inception-V3 | Inception-V3 · Caltech-256 | top-1 accuracy: 79.52% -> 80.61% | Improved |
| add WCD to Compact Bilinear CNN | Compact Bilinear CNN · CUB-200-2011 | top-1 accuracy: 84.01% -> 84.88% | Improved |
| add WCD (wrs 0.9, q 0.5) rather than Dropout after pool5 (ablation) | VGGNet-16 · CUB-200-2011 | top-1 accuracy: 73.90% -> 75.60% | Improved |
| add WCD rather than SE-Block after pool5 (ablation) | VGGNet-16 · CUB-200-2011 | top-1 accuracy: 73.05% -> 75.60% | Improved |
| use weighted random selection with RNG (q<1) instead of q=1 (ablation) | WCD on VGGNet-16 · CUB-200-2011 | top-1 accuracy: 73.28% -> 75.60% | Improved |
| use weighted selection (wrs<1) instead of Channel-wise Dropout (wrs=1) (ablation) | WCD on VGGNet-16 · CUB-200-2011 | top-1 accuracy: 75.02% -> 75.60% | Improved |
| SE-Block pretrained on ImageNet vs not (ablation) | VGGNet-16 · CUB-200-2011 | top-1 accuracy: 72.71% -> 73.05% | Improved |
| very low keep ratio (wrs 1 q 0.25 or wrs 0.25 q 1) (ablation) | WCD on VGGNet-16 · CUB-200-2011 | top-1 accuracy: 75.60% -> 73.16%; 72.16% | Hurt |
Closest papers by meaning
- DropBlock: A regularization method for convolutional networks (NeurIPS 2018 · similarity 0.87)
- Surprising properties of dropout in deep networks (JMLR 2018 · similarity 0.86)
- Shakeout: A New Regularized Deep Neural Network Training Scheme (AAAI 2016 · similarity 0.86)
- Dropout with Expectation-linear Regularization (ICLR 2017 · similarity 0.86)
- Shakeout: A New Approach to Regularized Deep Neural Network Training (TPAMI 2017 · similarity 0.86)
- Adaptive Dropout with Rademacher Complexity Regularization (ICLR 2018 · similarity 0.85)
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.
