hello

[chang556]

AttributeError: module ‘tensorflow‘ has no attribute ‘contrib‘

how can use ，because we are usu openvino2021.4.2

[microboym]

I recently succeeded in converting my model with openvino2021.4.

1. I created a separated virtual environment with anaconda, and run the following commands:

condo activate openvino-yolov4
conda install python==3.6.4
conda install pip
pip install tensorflow==1.15.5
pip install tqdm sklearn
python convert_weights_pb.py --class_names cfg/<your model>.names --weights_file <your model>.weights --data_format NHWC

which generated the frozen_darknet_yolov4_model.pb file.

2. I used my os python /usr/bin/python3 to create a venv, which was different from the condo env I had created, because OpenVINO seemed not capable with anaconda.

bash # use a different shell that has not been init with conda. I used zsh before so now I used bash.

/usr/bin/python3 -m pip install virtualenv
/usr/bin/python3 -m virtualenv -p `which python3` <directory_for_environment>
source <directory_for_environment>/bin/activate

# Than follow the installation guide provided by intel.
source /opt/intel/openvino_2021/bin/setupvars.sh
cd /opt/intel/openvino_2021/deployment_tools/model_optimizer/install_prerequisites
sudo ./install_prerequisites_tf.sh ## DO NOT use ./install_prerequisites.sh or ./install_prerequisites_tf2.sh

3. Finally, modify yolov4.json and run the mo script in the env created in step 2.

### Remember to use the same shell that you performed step2
python3 "/opt/intel/openvino_2021/deployment_tools/model_optimizer/mo.py" --input_model frozen_darknet_yolov4_model.pb --transformations_config yolov4.json --batch 1 --reverse_input_channels

Good luck