OpenVINO Model Optimizer not found when exporting object detection model

I am facing problems when running the export script in the object detection example here: https://github.com/openvinotoolkit/training_extensions/tree/develop/models/object_detection/model_templates/custom-object-detection

Although I am using an OpenVINO dev container, and already did the install_prerequisites.sh, the exporter cannot seem to convert the .pth file into a .bin and .xml file.

The warning I would get when running install_prerequisites is:

[WARNING] All Model Optimizer dependencies are installed globally.
[WARNING] If you want to keep Model Optimizer in separate sandbox
[WARNING] run install_prerequisites.sh "{caffe|tf|tf2|mxnet|kaldi|onnx}" venv

The warning I would get when running export.py is:

WARNING: ONNX Optimizer has been moved to https://github.com/onnx/optimizer.
All further enhancements and fixes to optimizers will be done in this new repo.
The optimizer code in onnx/onnx repo will be removed in 1.9 release.

Downloading /root/.torch/models/mobilenetv2_w1-0887-13a021bc.pth.zip from https://github.com/osmr/imgclsmob/releases/download/v0.0.213/mobilenetv2_w1-0887-13a021bc.pth.zip...
ONNX model has been saved to "/openvino_training_extensions/my_model/export/model.onnx"
OpenVINO Model Optimizer not found, please source openvino/bin/setupvars.sh before running this script.

The above claims that the Model Optimizer is not found.

To reproduce my problem, you can use this dockerfile, which assumes that I have another build step above named trained, containing all my training files:

FROM openvino/ubuntu18_dev:2021.3 
USER root
ENV DEBIAN_FRONTEND=noninteractive
COPY --from=train /openvino_training_extensions /openvino_training_extensions
ENV WORK_DIR='/openvino_training_extensions/my_model'
WORKDIR  ${WORK_DIR}
ENV OBJ_DET_DIR=/openvino_training_extensions/models/object_detection
ENV MODEL_TEMPLATE='./model_templates/custom-object-detection/mobilenet_v2-2s_ssd-256x256/template.yaml'
ENV TRAIN_ANN_FILE="${OBJ_DET_DIR}/../../data/airport/annotation_example_train.json"
ENV TRAIN_IMG_ROOT="${OBJ_DET_DIR}/../../data/airport/train"
ENV VAL_ANN_FILE="${OBJ_DET_DIR}/../../data/airport/annotation_example_val.json"
ENV VAL_IMG_ROOT="${OBJ_DET_DIR}/../../data/airport/val"
ENV CLASSES="vehicle,person,non-vehicle"
WORKDIR /openvino_training_extensions/models/object_detection
RUN apt-get update && apt-get install -y sudo 
ARG OTE_DIR='/openvino_training_extensions'
RUN ./init_venv.sh
RUN . venv/bin/activate && pip3 install -e /openvino_training_extensions/ote
RUN cd /openvino_training_extensions/models/object_detection &&\
   source /opt/intel/openvino/bin/setupvars.sh && \
   . venv/bin/activate && \
   /opt/intel/openvino_2021.3.394/deployment_tools/model_optimizer/install_prerequisites/install_prerequisites.sh && \
   cd /openvino_training_extensions/my_model && \
   python3 export.py \
   --load-weights ${WORK_DIR}/outputs/latest.pth \
   --save-model-to ${WORK_DIR}/export \

Environment:

• OS: Linux Ubuntu 18.04
• Framework version: PyTorch (as used in the custom-object-detector repo)
• Python version: 3.6.9 (as used in the container openvino/ubuntu18_dev:2021.3 on dockerhub)
• OpenVINO version: 2021.3
• CUDA/cuDNN version: -
• GPU model and memory: -

[pigubaoza]

The solution that ultimately worked for me was to use /opt/intel/openvino_2021.3.394/deployment_tools/model_optimizer/install_prerequisites/install_prerequisites.sh venv on the third last line. Note the use of venv at the end. I still used the virtual environments and concatenated operations as I wanted to follow the official steps as closely as possible

[pigubaoza]

For those using the Dockerhub container tagged openvino/ubuntu18_dev:2021.3 (I have not tested all of them), /opt/intel/openvino and /opt/intel/openvino_2021 symlinks to /opt/intel/openvino_2021.3.394/.

@ReshitkoM’s solution might work if you installed OpenVINO 2021 directly.