TensorFlow2.0-Tutorial

A TensorFlow2.0 tutorial for teaching purpose. (Course ended at Sep. 2019)

Dataset

Note that the folders for dataset and checkpoints are not uploaded in GitHub, so things may broke somewhere.

Temptatvie Schedule

1. Simple network
   • TensorFlow installation
   • Tensor and eager execution
   • Simple image classification example (ex. MINST)
   • Data preprocessing
   • Build model using tf.keras.Sequential
   • Inspect model and plot model graph
   • Keras callback
   • Save and load model
   • Add convolutional layers to model
   • Simple regression example (Auto MPG) (optional)
2. Keras functional API
   • Simple ResNet model
   • Building residual block
   • Complex graph topologies (ex. Image Colorization)
   • Model with shared layers
   • Model with multiple inputs and ouputs
   • Nice coding practices
3. Data input pipeline using TensorFlow Dataset
   • tf.data input pipeline (ex. AOI image classification)
   • Train/validation split
   • Data augmentation
   • Write loss and metrics to csv file
   • Train model on cloud virtual machine
   • Short guide on GPU choice
4. Transfer learning
   • Transfer learning with pretrained CNN
5. Generative models
   • DCGAN (Generating anime faces using Getchu dataset)
   • Constructing data input pipeline from TFRecord
   • Custom training loop with tf.GradientTape()

Contact

Please file an issue if there's any error, or suggest better coding practice. Thanks!