Move examples/ to models/.

Add pre-release version of code for the paper "High-Fidelity Generative Image Compression".

PiperOrigin-RevId: 317378744
Change-Id: Id4ebf79f8b87da2fa81c24c5d9c237e106c42fa2

Author: mentzer

BUILD

@@ -54,21 +54,21 @@ py_binary(
 
 py_binary(
     name = "tfci",
-    srcs = ["examples/tfci.py"],
+    srcs = ["models/tfci.py"],
     python_version = "PY3",
     deps = [":tensorflow_compression"],
 )
 
 py_binary(
     name = "bls2017",
-    srcs = ["examples/bls2017.py"],
+    srcs = ["models/bls2017.py"],
     python_version = "PY3",
     deps = [":tensorflow_compression"],
 )
 
 py_binary(
     name = "bmshj2018",
-    srcs = ["examples/bmshj2018.py"],
+    srcs = ["models/bmshj2018.py"],
     python_version = "PY3",
     deps = [":tensorflow_compression"],
 )

README.md

@@ -119,7 +119,7 @@ import tensorflow_compression as tfc
 ### Using a pre-trained model to compress an image
 
 In the
-[examples directory](https://github.com/tensorflow/compression/tree/master/examples),
+[models directory](https://github.com/tensorflow/compression/tree/master/models),
 you'll find a python script `tfci.py`. Download the file and run:
 ```bash
 python tfci.py -h
@@ -142,7 +142,7 @@ appended (any existing extensions will not be removed).
 ### Training your own model
 
 The
-[examples directory](https://github.com/tensorflow/compression/tree/master/examples)
+[models directory](https://github.com/tensorflow/compression/tree/master/models)
 contains an implementation of the image compression model described in:
 
 > "End-to-end optimized image compression"<br />

examples/bls2017.py renamed to models/bls2017.py

@@ -0,0 +1,101 @@
+# High-Fidelity Generative Image Compression
+
+<div align="center">
+  <a href='https://hific.github.io'>
+  <img src='https://hific.github.io/social/thumb.jpg' width="80%"/>
+  </a>
+</div>
+
+## [[Demo]](https://hific.github.io) [[Paper]](https://arxiv.org/abs/2006.09965) [[Colab]](https://colab.research.google.com/github/tensorflow/compression/blob/master/models/hific/colab.ipynb)
+
+
+## Abstract
+
+We extensively study how to combine Generative Adversarial Networks and learned
+compression to obtain a state-of-the-art generative lossy compression system. In
+particular, we investigate normalization layers, generator and discriminator
+architectures, training strategies, as well as perceptual losses. In contrast to
+previous work, i) we obtain visually pleasing reconstructions that are
+perceptually similar to the input, ii) we operate in a broad range of bitrates,
+and iii) our approach can be applied to high-resolution images. We bridge the
+gap between rate-distortion-perception theory and practice by evaluating our
+approach both quantitatively with various perceptual metrics and a user study.
+The study shows that our method is preferred to previous approaches even if they
+use more than 2&times; the bitrate.
+
+## Try it out!
+
+[<img src="https://colab.research.google.com/assets/colab-badge.svg" align="center">](https://colab.research.google.com/github/tensorflow/compression/blob/master/models/hific/colab.ipynb)
+
+We show some images on the [demo page](https://hific.github.io) and we
+release a
+[colab](https://colab.research.google.com/github/tensorflow/compression/blob/master/models/hific/colab.ipynb)
+update for interactively using our models on your own images.
+
+## Using the code
+
+In addition to `tensorflow_compression`, you need to install [`compare_gan`](https://github.com/google/compare_gan)
+and TensorFlow 1.15:
+
+```bash
+pip install -r requirements.txt
+```
+
+## Running our models locally
+
+Use `tfci.py` for locally running our models to encode and decode images:
+
+```python
+python tfci.py compress <model> <PNG file>
+```
+
+where `model` can be one of `"hific-lo", "hific-mi", "hific-hi"`.
+
+## Code
+
+The architecture is defined in `arch.py` , which is used to build the model in
+`model.py`. Our configurations are in `configs.py`.
+
+### Training your own models.
+
+We release a _simplified_ trainer in `train.py` as a starting point for custom
+training. Note that it's using [LSUN]() from [tfds]() which likely needs to be
+adapted to a bigger dataset to obtain state-of-the-art results (see below).
+
+For the paper, we initialize our GAN models from a MSE+LPIPS checkpoint. To
+replicate this, first train a model for MSE + LPIPS only, and then use that as a
+starting point:
+
+```bash
+# First train a model for MSE+LPIPS:
+python train.py --config mselpips --ckpt_dir ckpts --num_steps 1M
+
+# Once that finishes, train a GAN model:
+python train.py --config hific --ckpt_dir ckpts \
+                --init_from ckpts/mselpips --num_steps 1M
+```
+
+To test a trained model, use `eval.py`:
+
+```bash
+python eval.py --config hific --ckpt_dir ckpts/hific
+```
+
+#### Adapting the dataset
+
+You can change to any other TFDS dataset by changing the `tfds_name` flag for
+`build_input`. To train on a custom dataset, you can replace the `_get_dataset`
+call in `train.py`.
+
+## Citation
+
+If you use the work released here for your research, please cite this paper:
+
+```
+@inproceedings{mentzer2020hific,
+  title={High-Fidelity Generative Image Compression},
+  author={Fabian Mentzer and George Toderici and Michael Tschannen and Eirikur Agustsson},
+  year={2020}
+}
+```
+