@@ -38,7 +38,7 @@ Use `tfci.py` for locally running our models to encode and decode images:
38
38
39
39
``` bash
40
40
git clone https://github.com/tensorflow/compression
41
- cd compression/compression/ models
41
+ cd compression/models
42
42
python tfci.py compress < model> < PNG file>
43
43
```
44
44
@@ -49,22 +49,19 @@ where `model` can be one of `"hific-lo", "hific-mi", "hific-hi"`.
49
49
50
50
## Using the code
51
51
52
+ To use the code, create a conda environment using Python 3.7, and the packages
53
+ listed in
54
+ [ requirements.txt] ( https://github.com/tensorflow/compression/blob/master/models/hific/requirements.txt ) .
52
55
53
- To use the code, create a conda environment using Python 3.6
54
- (newer is not supported at the moment), and the following packages.
55
-
56
- ** NOTE** : We only support CUDA 10.0, Python 3.6, and TensorFlow 1.15.
57
- TensorFlow must be installed via pip, not conda.
58
- Any other setup is not going to work (we tested newer versions of Tensorflow
59
- and Python and they don't work). We're working on a fix.
56
+ ** NOTE** : At the moment, we only support CUDA 10.0, Python 3.6-3.7, TensorFlow
57
+ 1.15, and Tensorflow Compression 1.3. TensorFlow must be installed via pip, not
58
+ conda. Unfortunately, newer versions of Tensorflow or Python will not work due
59
+ to various constraints in the dependencies and in the TF binary API.
60
60
61
61
``` bash
62
- conda create --name hific python=3.6 cudatoolkit=10.0 cudnn
62
+ conda create --name hific python=3.7 cudatoolkit=10.0 cudnn
63
63
conda activate hific
64
- pip install tensorflow-gpu==1.15 # Make sure to install TF via pip, not conda!
65
- pip install git+git://github.com/google/compare_gan@19922d3004b675c1a49c4d7515c06f6f75acdcc8
66
- pip install tensorflow-compression==1.3
67
- pip install Pillow
64
+ pip install -r hific/requirements.txt
68
65
```
69
66
70
67
#### Note on CUDNN Errors
@@ -115,7 +112,7 @@ python -m hific.train --config mselpips --ckpt_dir ckpts/mse_lpips \
115
112
116
113
# Once that finishes, train a GAN model:
117
114
python -m hific.train --config hific --ckpt_dir ckpts/hific \
118
- --init_autoencoder_from_ckpt_dir ckpts/mselpips \
115
+ --init_autoencoder_from_ckpt_dir ckpts/mse_lpips \
119
116
--num_steps 1M \
120
117
--tfds_dataset_name coco2014
121
118
```
0 commit comments