TensorFlow 2.0 has no attribute 'ConfigProto'; fix proposed

[etjones]

Nice project! Thanks for putting it together!

I found a breaking issue today, with a simple fix. The versioning strings ask for TensorFlow >=1.14, but there's some internal code that relies on a TF configuration object that was changed in TF2.0. A workaround is to change the tensorflow requirement strings in setup.py and requirements.txt from "tensorflow>=1.14", to "tensorflow>=1.14,<2.0",

I did a pip install fer in a new virtualenv and then started following the usage instructions. Here's a session dump:

In [1]: import tensorflow    
In [2]: tensorflow.__version__  
Out[2]: '2.0.0'
In [3]: from fer import FER   
In [4]: detector = FER()      
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-2-a308cfd708c0> in <module>
----> 1 detector = FER()

~/td_venv/lib/python3.7/site-packages/fer/fer.py in __init__(self, cascade_file, mtcnn, emotion_model, scale_factor, min_face_size, min_neighbors, offsets, compile)
    110                 "fer", "data/emotion_model.hdf5"
    111             )
--> 112             self.config = tf.ConfigProto(log_device_placement=False)
    113             self.config.gpu_options.allow_growth = True
    114 

AttributeError: module 'tensorflow' has no attribute 'ConfigProto'

Here's a workaround for TF1/TF2 compatibility that I think would also resolve things.

I'll send a PR later on if I can get something working smoothly. Cheers!

[JustinShenk]

Hi Evan, thanks for your message and proposed fix. I would be happy to review a PR. Cheers, Justin

…

On Fri, Nov 1, 2019 at 11:23 PM Evan Jones ***@***.***> wrote: Nice project! Thanks for putting it together! I found a breaking issue today, with a simple fix. The versioning strings ask for TensorFlow >=1.14, but there's some internal code that relies on a TF configuration object that was changed in TF2.0. A workaround is to change the tensorflow requirement strings in setup.py and requirements.txt from "tensorflow>=1.14", to "tensorflow>=1.14,<2.0", I did a pip install fer in a new virtualenv and then started following the usage instructions. Here's a session dump: In [1]: import tensorflow In [2]: tensorflow.__version__ Out[2]: '2.0.0' In [3]: from fer import FER In [4]: detector = FER() --------------------------------------------------------------------------- AttributeError Traceback (most recent call last)<ipython-input-2-a308cfd708c0> in <module> ----> 1 detector = FER() ~/td_venv/lib/python3.7/site-packages/fer/fer.py in __init__(self, cascade_file, mtcnn, emotion_model, scale_factor, min_face_size, min_neighbors, offsets, compile) 110 "fer", "data/emotion_model.hdf5" 111 ) --> 112 self.config = tf.ConfigProto(log_device_placement=False) 113 self.config.gpu_options.allow_growth = True 114 AttributeError: module 'tensorflow' has no attribute 'ConfigProto' Here's a workaround <https://stackoverflow.com/a/56994817> for TF1/TF2 compatibility that I think would also resolve things. I'll send a PR later on if I can get something working smoothly. Cheers! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#7>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOLMZFFMUZA37B3FJ7MRZDQRST4ZANCNFSM4JIAUZDQ> .

[helenou]

Hi @JustinShenk I started using fer and had this issue. I did install from pip3 as I have python 3.6
Finally to get the fix from master i cloned this repo manually and that let me start.

No idea why pip isn't picking the right one, just wanted to let you know error still showed then

[JustinShenk]

Fixed in v20.0.0 with dc46da3. Now works with pip install -U fer.