Clone models repostory
git clone https://github.com/tensorflow/models.git
Compile Proto Buffer (protobof) inside models/research
directory
protoc object_detection/protos/*.proto --python_out=.
And then export $PYTHONPATH variable inside models/research
directory
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
Setuptools
sudo python3 setup.py install
Convert dataset meta labels xml to csv
python3 xml_to_csv.py
Convert csv dataset labels to TFRecord file format
python3 generate_tfrecord.py --type=train --csv_input=data/train_labels.csv --output_path=data/train.record
python3 generate_tfrecord.py --type=test --csv_input=data/test_labels.csv --output_path=data/test.record
Grab the COCO models
wget http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz
or you can click here
Start training the model
python3 train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_pets.config
Export graph and the model we have trained
python3 export_inference_graph.py \
--input_type image_tensor \
--pipeline_config_path training/ssd_mobilenet_v1_pets.config \
--trained_checkpoint_prefix training/model.ckpt-1000 \
--output_directory plate_model_exported
Start jupyter notebook inside the directory to see the results
jupyter notebook --notebook-dir=.