Based on this tutorial
If you want to see the inference (execution): https://github.com/Coofy/sight-bot
Miniconda:
conda create -n perception-tf1.15 pip python=3.7 pyqt=5
pip install tensorflow==1.15
cd addons/labelImg
sudo apt-get install pyqt5-dev-tools
sudo pip install -r requirements/requirements-linux-python3.txt
make qt5py3
git clone --depth 1 https://github.com/tensorflow/models
cd models/research/
protoc object_detection/protos/*.proto --python_out=.
pip install .
cd workspace/training_demo
TF_ENABLE_GPU_GARBAGE_COLLECTION=false TF_FORCE_GPU_ALLOW_GROWTH=true python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/faster_rcnn_inception_v2_coco.config