Notes for installation in 2023...

[forever208]

Since most people are using Pytorch or Tensorflow 2 and Python3, installing this tensorflow1.4 repo is difficult for people who want to use it in 2023. I first provide the full installation below

Ubuntu 20.04:

1. create a virtual env (make sure to specify python=2.7, otherwise you can not install TensorFlow 1.x using pip)

conda create -n PRNet pyton=2.7 

2. install TensorFlow 1.6

pip install tensorflow==1.6

3. install dlib

pip install dlib

4. install other libs

pip install scipy
pip install scikit-image

RIght now if you run demo.py and have this issue Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA. Congratulations, you need to build the TensorFlow on your own (refer to stack overflow for the reason).

As for building TensorFlow 1.6 on your own, first go to this github , under the list Expand for older builds, download the 1.6.0 tensorflow cpu whl file. Then you build by:

pip install --ignore-installed --upgrade /PATH/TO/BINARY.whl --user

Since TensorFlow 1.6 does not support the latest CUDA version, I can not find a way to use TensorFlow-gpu 1.6 right now

[yumikim381]

Thank you so much for your guide
After building TensorFlow 1.6 with the command you've given,
i still get the error :

Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX512F.

Does anyone else also have this error? I find it wierd, cause i m trying to run same code as everyone else here and it seems to use AVX512F

[YacratesWyh]

cannot install dlib for this version, wonder any recent pr

[xubaixinxbx]

cannot install dlib for this version, wonder any recent pr

Hi, the same question, it seems python with version >= 3.6 could pip dlib successfully done

[mwb3262716541]

Just add "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2' " to the run_basics.py file. Ignore AVX support

[DiaosWang]

cannot install dlib for this version, wonder any recent pr

i 'try conda install dlib', and success