https://github.com/lukasturcani/dnn_tensorflow
Implementations of deep network architectures in TensorFlow.
https://github.com/lukasturcani/dnn_tensorflow
Last synced: 7 months ago
JSON representation
Implementations of deep network architectures in TensorFlow.
- Host: GitHub
- URL: https://github.com/lukasturcani/dnn_tensorflow
- Owner: lukasturcani
- Created: 2019-05-22T08:47:33.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2019-05-22T10:23:31.000Z (over 6 years ago)
- Last Synced: 2025-01-10T17:17:45.646Z (9 months ago)
- Language: Python
- Homepage:
- Size: 66.4 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: readme.rst
Awesome Lists containing this project
README
:author: Lukas Turcani
I no longer maintain this repo, and the code in it is out of date.
For example the user of TensorFlow layers should be replaced with
the TensorFlow keras api. I maintain
https://www.github.com/lukasturcani/dnn instead.
This repo provides a number of network architectures, implemented in
TensorFlow as well as scripts used to run them.
To train a network, run the appropriate run script. For example::
$ python -m dnn_tensorflow.train_scripts.cifar100.simple_cnn
Notice that the scripts are run from the module level using the
``python -m`` flag.
Networks can be modified using command line arguments, for example::
$ python -m dnn_tensorflow.train_scripts.mnist.simple_cnn --conv_in_channels 1 20 50 --conv_out_channels 20 50 60 --conv_kernel_size 5 5 5 --conv_strides 1 1 1 --conv_paddings 0 0 0 --conv_dilations 1 1 1 --pool_kernel_sizes 2 2 2 --pool_strides 2 2 2 --pool_paddings 0 0 0 --pool-dilations 1 1 1 --train_batch_size 100 --label_smoothing 0.5 --epochs 10
changes the default values for training batch size, label smoothing and
the number of training epochs and adds a new convolution layer and pooling
layer vs the default network. Any number of layers can be added / removed
via the command line in this fashion.
Each script can have options viewed by::
$ python -m path.to.train.script --help
Results
=======
This is a short summary of some of the nice results from this repo.
Not all implemented architectures are listed here.
* `Image Inpainting`_
* `DCGAN`_
* `FCGAN`_
Image Inpainting
----------------
This is a task where the generator is provided with an image that
is missing some pixels and it is asked to fill them in. In this
example, I cover up the bottom half of MNIST images and get the
generator to fill them in.
TensorFlow implementation
.........................
Run with::
$ python -m dnn.tensorflow.train_scripts.mnist.image_inpainting
Results:
.. image::
DCGAN
-----
A more advanced GAN architecture, which is fully convolutional.
Run with::
$ python -m dnn_tensorflow.train_scripts.mnist.dcgan
To generate results::
$ python -m dnn_tensorflow.train_scripts.mnist.dcgan --sample_generator
.. image:: images/mnist_dcgan.jpg
DCGAN with labels.
..................
This network can also be trained using the labelling information
present in the MNIST dataset. This leads to better results and each
digit can be sampled explicitly.
Run with::
$ python -m dnn.tensorflow.train_scripts.mnist.dcgan --labels
To generate results::
$ python -m dnn.tensorflow.train_scripts.mnist.dcgan --labels --sample_generator
.. image:: images/mnist_dcgan_labels.jpg
FCGAN
-----
This is a vanilla GAN using feed-forward networks as both the
generator and discriminator.
Run with::
$ python -m dnn_tensorflow.train_scripts.mnist.fcgan
To generate results::
$ python -m dnn_tensorflow.train_scripts.mnist.fcgan --sample_generator
.. image:: images/mnist_fcgan.jpg
FCGAN with labels.
..................
This network can also be trained using the labelling information
present in the MNIST dataset. This leads to better results and each
digit can be sampled explicitly.
Run with::
$ python -m dnn_tensorflow.train_scripts.mnist.dcgan --labels
To generate results::
$ python -m dnn_tensorflow.train_scripts.mnist.dcgan --labels --sample_generator
.. image:: images/mnist_fcgan_labels.jpg