resnets-0.0.1.tar.gz

# Resnets Tidy implementation of classical neural networks for classification. Available frameworks: * Jax * PyTorch * TensorFlow Everything in one place with results matching those reported in papers. [](https://colab.research.google.com/github/gahaalt/cifar-vs-tensorflow2/blob/master/Playground.ipynb) ## Installation Install the package with pip: ```bash pip install resnets ``` ## Available architectures ### VGG | architecture | parameters | reported best | this repository | CIFAR-10 ckpt | ImageNet ckpt | |-------------:|:----------:|:-------------:|:---------------:|---------------|---------------| | VGG11 | 9.2M | 7.81 | 7.98 | | | | VGG13 | 9.4M | 6.35 | 6.17 | | | | VGG16 | 14.7M | 6.49 | 6.34 | | | | VGG19 | 20.0M | 6.76 | 6.72 | | | Sources: * [Very Deep Convolutional Network for Large-Scale Image Recognition](https://arxiv.org/abs/1409.1556) * [On Correlation of Features Extracted by Deep Neural Networks](https://arxiv.org/abs/1901.10900) ### ResNet | architecture | parameters | reported best | this repository | CIFAR-10 ckpt | ImageNet ckpt | |-------------:|:----------:|:-------------:|:---------------:|---------------|---------------| | ResNet-20 | 0.27M | 8.75 | 7.99 | | | | ResNet-32 | 0.46M | 7.51 | 7.40 | | | | ResNet-44 | 0.66M | 7.17 | 6.83 | | | | ResNet-56 | 0.85M | 6.97 | 6.23 | | | | ResNet-110 | 1.7M | 6.37 | 5.98 | | | | ResNet-164 | 1.7M | 5.46 | 5.27 | | | | ResNet-1001 | 10.3M | 4.92 | 5.06 | | | Sources: * [Deep Residual Learning for Image Recognition](https://arxiv.org/abs/1512.03385) * [Identity Mappings in Deep Residual Networks](https://arxiv.org/abs/1603.05027) ### Wide ResNet | architecture | parameters | reported mean | this repository | CIFAR-10 ckpt | ImageNet ckpt | |-------------:|:----------:|:-------------:|:---------------:|---------------|---------------| | WRN-16-4 | 2.7M | 5.02 | | | | | WRN-40-4 | 8.9M | 4.53 | 4.46 | | | | WRN-16-8 | 11.0M | 4.27 | | | | | WRN-28-10 | 36.5M | 4.00 | | | | | WRN-28-10+ | 36.5M | 3.89 | | | | (+ with dropout) Sources: * [Wide Residual Networks](https://arxiv.org/abs/1605.07146) ### ResNeXt