README.md
1 # GAN with TensorFlow eager execution
2
3 A simple Generative Adversarial Network (GAN) example using eager execution.
4 The discriminator and generator networks each contain a few convolution and
5 fully connected layers.
6
7 Other eager execution examples can be found under the parent directory.
8
9 ## Content
10
11 - `mnist.py`: Model definitions and training routines.
12 - `mnist_test.py`: Benchmarks for training and using the models using eager
13 execution.
14 - `mnist_graph_test.py`: Benchmarks for training and using the models using
15 graph execution. The same model definitions and loss functions are used in
16 all benchmarks.
17
18
19 ## To run
20
21 - Make sure you have installed TensorFlow 1.5+ or the latest `tf-nightly`
22 or `tf-nightly-gpu` pip package in order to access the eager execution feature.
23
24 - Train model. E.g.,
25
26 ```bash
27 python mnist.py
28 ```
29
30 Use `--output_dir=<DIR>` to direct the script to save TensorBoard summaries
31 during training. Disabled by default.
32
33 Use `--checkpoint_dir=<DIR>` to direct the script to save checkpoints to
34 `<DIR>` during training. DIR defaults to /tmp/tensorflow/mnist/checkpoints/.
35 The script will load the latest saved checkpoint from this directory if
36 one exists.
37
38 Use `-h` for other options.
39