Research map / Architectures and efficiency / CNN design
Network Deconvolution
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 |
|---|---|---|---|
| Replace batch normalization with network deconvolution | VGG-16 · CIFAR-10 | accuracy (100 epochs): 93.58% -> 94.56% | Helped |
| Replace batch normalization with network deconvolution | VGG-16 · CIFAR-100 | accuracy (100 epochs): 72.75% -> 75.32% | Helped |
| Replace batch normalization with network deconvolution | VGG-11 · ImageNet | top-1 accuracy: 70.38% -> 71.95% | Helped |
| Replace batch normalization with network deconvolution | VGG-11 · ImageNet | top-5 accuracy: 89.81% -> 90.49% | Helped |
| Replace batch normalization with network deconvolution | ResNet-18 · ImageNet | top-1 accuracy: 69.76% -> 71.24% | Helped |
| Replace batch normalization with network deconvolution | ResNet-18 · ImageNet | top-5 accuracy: 89.08% -> 90.14% | Helped |
| Replace batch normalization with network deconvolution | DenseNet-121 · ImageNet | top-1 accuracy: 74.65% -> 75.73% | Helped |
| Replace batch normalization with network deconvolution | DenseNet-121 · ImageNet | top-5 accuracy: 92.17% -> 92.75% | Helped |
| Replace batch normalization with network deconvolution | 10 modern CNN architectures · CIFAR-10 and CIFAR-100 | accuracy | Helped |
| Use network deconvolution instead of vanilla SGD normalization-free training | Fashion-MNIST linear regression · Fashion-MNIST | convergence iterations: ~ 5 iterations | Helped |
| Increase batch size while adjusting learning rate, regularization, and iteration settings (ablation) | Network deconvolution · Unspecified accuracy benchmark | accuracy: 89.12% to 91.66% | Mixed |
| Use stronger weight decay with network deconvolution rather than batch normalization (ablation) | VGG-13 · CIFAR-100 | accuracy | Mixed |
Closest papers by meaning
- Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks (ICLR 2019 · similarity 0.89)
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · similarity 0.89)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.89)
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.89)
- Involution: Inverting the Inherence of Convolution for Visual Recognition (CVPR 2021 · similarity 0.88)
- Omni-Kernel Network for Image Restoration (AAAI 2024 · similarity 0.87)
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.