Apress Source Code

This repository contains the source code/demos for the book Convolutional Neural Networks with Swift for TensorFlow by Brett Koonce (Apress, 2021).

Download the files as a zip using the green button, or clone the repository to your machine using Git.

Releases

Release v1.0 corresponds to the code in the published book, without corrections or updates.

To run it locally, you will need a working swift for tensorflow installation.

Quickstart

Set up GCP.

export ZONE="us-central1-c"
export GPU_TYPE="t4"

gcloud compute instances create s4tf-ubuntu-${GPU_TYPE} \
--zone=${ZONE} \
--image-project=deeplearning-platform-release \
--image-family=swift-latest-gpu-ubuntu-1804 \
--maintenance-policy=TERMINATE \
--accelerator="type=nvidia-tesla-${GPU_TYPE},count=1" \
--metadata="install-nvidia-driver=True" \
--machine-type=n1-highmem-2 \
--boot-disk-size=256GB


git clone https://github.com/Apress/convolutional-neural-networks-with-swift-for-tensorflow.git
cd convolutional-neural-networks-with-swift-for-tensorflow

$ swift run
error: multiple executable products available: VGG, SqueezeNet, Resnet50, Resnet34, MobileNetV3, MobileNetV2, MobileNetV1, MNIST-XLA-TPU, MNIST-2D, MNIST-1D, EfficientNet, CIFAR

Now you can run the individual demos, eg swift run MNIST-1D.

Outline

The book is structured as follows:

Basic

We will explore the basic building blocks of neural networks and how to combine them with convolutions to perform simple image recognition tasks.

• MNIST-1D
• MNIST-2D
• CIFAR

Advanced

We will build upon the above to produce actual state of the art approaches in this field.

• VGG
• Resnet34
• Resnet50

Mobile

We will look at some different approaches for mobile devices, which require us to utilize our computing resources carefully.

• SqueezeNet
• MobileNetV1
• MobileNetV2

State of the art

We look at the work that leads up to EfficientNet, the current state of art for image recognition.

• EfficientNet
• MobileNetV3

Future

We look at how swift for tensorflow's use of XLA internally allows our code to scale to large supercomputer clusters.

• MNIST-XLA-TPU

Contributions

See the file Contributing.md for more information on how you can contribute to this repository.