1- # TensorFlow Compression
1+ # Introduction
22
33This project contains data compression ops and layers for TensorFlow. The
44project website is at
@@ -37,54 +37,44 @@ There's also an introduction to our `EntropyBottleneck` class
3737and a description of the range coding operators
3838[ here] ( https://tensorflow.github.io/compression/docs/range_coding.html ) .
3939
40- ## Quick start
40+ ## Installation
4141
42- ### Installing release 1.1 (stable)
43-
44- Install TensorFlow 1.13 using one of the methods described in the
45- [ TensorFlow installation instructions] ( https://www.tensorflow.org/install ) .
46-
47- Download the ZIP file for
48- [ release v1.1] ( https://github.com/tensorflow/compression/releases/tag/v1.1 )
49- and unpack it. Then include its root directory in your ` PYTHONPATH `
50- environment variable:
51-
52- ``` bash
53- cd < target directory>
54- wget https://github.com/tensorflow/compression/archive/v1.1.zip
55- unzip v1.1.zip
56- export PYTHONPATH=" $PWD /compression-1.1:$PYTHONPATH "
57- ```
58-
59- To make sure the library imports succeed, try running the unit tests:
60-
61- ``` bash
62- cd compression-1.1
63- for i in tensorflow_compression/python/* /* _test.py; do python $i ; done
64- ```
65-
66- ### Installing release 1.2b2 (beta)
42+ *** Note: Precompiled packages are currently only provided for Linux and Darwin
43+ (Mac OS). To use these packages on Windows, consider using a [ TensorFlow
44+ Docker image] ( https://www.tensorflow.org/install/docker ) and installing
45+ tensorflow-compression using pip inside the Docker container.***
6746
6847Set up an environment in which you can install precompiled binary Python
6948packages using the ` pip ` command. Refer to the
7049[ TensorFlow installation instructions] ( https://www.tensorflow.org/install/pip )
7150for more information on how to set up such a Python environment.
7251
73- Install TensorFlow 1.14 or above with or without GPU support:
52+ The current version of tensorflow-compression requires TensorFlow 1.14 or
53+ above. You can install TensorFlow from any source. To install it via ` pip ` , run
54+ the following command:
7455``` bash
75- pip install tensorflow
56+ pip install tensorflow-gpu
7657```
77- or
58+ for GPU support, or
7859``` bash
79- pip install tensorflow-gpu
60+ pip install tensorflow
8061```
62+ for CPU-only.
8163
82- Then, run the following command to install the binary PIP package:
64+ Then, run the following command to install the tensorflow-compression ` pip `
65+ package:
8366``` bash
8467pip install tensorflow-compression
8568```
8669
87- ### Using the library
70+ To test that the installation works correctly, you can run the unit tests with
71+ ``` bash
72+ python -m tensorflow_compression.python.all_test
73+ ```
74+ Once the command finishes, you should see a message ``` OK (skipped=11) ``` or
75+ similar in the last line.
76+
77+ ## Usage
8878
8979We recommend importing the library from your Python code as follows:
9080
@@ -95,31 +85,23 @@ import tensorflow_compression as tfc
9585
9686### Using a pre-trained model to compress an image
9787
98- *** Note: you need to have a release >1.1 installed for pre-trained model
99- support.***
100-
10188In the
10289[ examples directory] ( https://github.com/tensorflow/compression/tree/master/examples ) ,
10390you'll find a python script ` tfci.py ` . Download the file and run:
104-
10591``` bash
10692python tfci.py -h
10793```
10894
10995This will give you a list of options. Briefly, the command
110-
11196``` bash
11297python tfci.py compress < model> < PNG file>
11398```
114-
11599will compress an image using a pre-trained model and write a file ending in
116100` .tfci ` . Execute ` python tfci.py models ` to give you a list of supported
117101pre-trained models. The command
118-
119102``` bash
120103python tfci.py decompress < TFCI file>
121104```
122-
123105will decompress a TFCI file and write a PNG file. By default, an output file
124106will be named like the input file, only with the appropriate file extension
125107appended (any existing extensions will not be removed).
@@ -135,15 +117,13 @@ contains an implementation of the image compression model described in:
135117> https://arxiv.org/abs/1611.01704
136118
137119To see a list of options, download the file ` bls2017.py ` and run:
138-
139120``` bash
140121python bls2017.py -h
141122```
142123
143124To train the model, you need to supply it with a dataset of RGB training images.
144125They should be provided in PNG format. Training can be as simple as the
145126following command:
146-
147127``` bash
148128python bls2017.py -v --train_glob=" images/*.png"
149129```
@@ -163,34 +143,34 @@ enough (or larger). This is described in more detail in:
163143If you wish, you can monitor progress with Tensorboard. To do this, create a
164144Tensorboard instance in the background before starting the training, then point
165145your web browser to [ port 6006 on your machine] ( http://localhost:6006 ) :
166-
167146``` bash
168147tensorboard --logdir=. &
169148```
170149
171150When training has finished, the Python script can be used to compress and
172151decompress images as follows. The same model checkpoint must be accessible to
173152both commands.
174-
175153``` bash
176154python bls2017.py [options] compress original.png compressed.bin
177155python bls2017.py [options] decompress compressed.bin reconstruction.png
178156```
179157
180- ## Building PIP packages
158+ ## Building ` pip ` packages
181159
182- This section describes steps to build PIP packages.
160+ This section describes the necessary steps to build your own ` pip ` packages of
161+ tensorflow-compression. This may be necessary to install it on platforms for
162+ which we don't provide precompiled binaries (currently only Linux and Darwin).
183163
184- We use the Docker image ` tensorflow/tensorflow:nightly-custom-op ` for building
185- PIP packages. Note that this is different from ` tensorflow/tensorflow:devel ` . To
186- be compatible with the TensorFlow PIP package, the GCC version must match, but
187- ` tensorflow/tensorflow:devel ` has a different GCC version installed.
164+ We use the Docker image ` tensorflow/tensorflow:nightly-custom-op ` for building
165+ ` pip ` packages. Note that this is different from ` tensorflow/tensorflow:devel ` .
166+ To be compatible with the TensorFlow ` pip ` package, the GCC version must match,
167+ but ` tensorflow/tensorflow:devel ` has a different GCC version installed.
188168
189- Inside a Docker container from the image, the following steps need to be done .
169+ Inside a Docker container from the image, the following steps need to be taken .
190170
191- 1 . Install TensorFlow PIP package.
192- 2 . Clone ` tensorflow- compression ` repo from GitHub.
193- 3 . Run ` :build_pip_pkg ` inside the cloned repo.
171+ 1 . Install TensorFlow ` pip ` package.
172+ 2 . Clone the ` tensorflow/ compression ` repo from GitHub.
173+ 3 . Run ` :build_pip_pkg ` inside the cloned repo:
194174
195175``` bash
196176sudo docker run -v /tmp/tensorflow_compression:/tmp/tensorflow_compression \
@@ -204,28 +184,22 @@ sudo docker run -v /tmp/tensorflow_compression:/tmp/tensorflow_compression \
204184The wheel file is created inside ` /tmp/tensorflow_compression ` . Optimization
205185flags can be passed via ` --copt ` to the ` bazel run ` command above.
206186
207- ### Testing PIP package
208-
209- First install the built wheel file.
210-
187+ To test the created package, first install the resulting wheel file:
211188``` bash
212189pip install /tmp/tensorflow_compression/tensorflow_compression-* .whl
213190```
214191
215- And run tests. (Do not run the tests in the workspace directory where
192+ Then run the unit tests (Do not run the tests in the workspace directory where
216193` WORKSPACE ` of ` tensorflow_compression ` repo lives. In that case, the Python
217- interpreter attempts to import ` tensorflow_compression ` packages from the source
218- tree rather than from the installed package system directory.)
219-
194+ interpreter would attempt to import ` tensorflow_compression ` packages from the
195+ source tree rather than from the installed package system directory):
220196``` bash
221197pushd /tmp
222198python -m tensorflow_compression.python.all_test
223199popd
224200```
225201
226- When done, uninstall the PIP package. The package name is
227- ` tensorflow-compression ` with a hyphen(-).
228-
202+ When done, you can uninstall the ` pip ` package again:
229203``` bash
230204pip uninstall tensorflow-compression
231205```
0 commit comments