jsonw99/object_detect_serving

0. download the pretrained model

cd objectDetectServing/object_detection
wget http://download.tensorflow.org/models/object_detection/ssd_inception_v2_coco_11_06_2017.tar.gz
tar xvzf ssd_inception_v2_coco_11_06_2017.tar.gz
rm ssd_inception_v2_coco_11_06_2017.tar.gz

1. setup environment

cd ..
protoc object_detection/protos/*.proto --python_out=.
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

2. build pretrained model to server

python object_detection/my_export_inference_graph.py \
--input_type=image_tensor \
--pipeline_config_path=object_detection/samples/configs/ssd_inception_v2_coco.config \
--trained_checkpoint_prefix=object_detection/ssd_inception_v2_coco_11_06_2017/model.ckpt \
--output_directory /tmp/inception_v2_model/ \
--model_version 1

3. setup server

change the directory to the folder of tensorflow_serving

cd ~
cd serving
bazel-bin/tensorflow_serving/model_servers/tensorflow_model_server --port=9000 --model_name=inception_v2 --model_base_path=/tmp/inception_v2_model/ --enable_batching=true

4. test

open another terminal

cd objectDetectServing

protoc object_detection/protos/*.proto --python_out=.
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

python object_detection/my_client.py --server=localhost:9000 --image=object_detection/test_images/image1.jpg	

5. notes

the directories "object_detection" and "slim" are from the tensorflow repository "tensorflow/models/research/".
the files "my_client.py", "my_export_inference_graph.py", and "my_exporter.py" are the original codes or modified codes.