This document is to give step by step guidance to allow the newer install and quickly familiar with OpenVINO.
OS/ENV
- · Recommend OS is Ubuntu 16.04 or Ubuntu18.04. The Centos or Windows will have some trouble for Installation.
- · This guide provides the steps for creating a Docker* image for Ubuntu system and further installation.
- · Set python to python3 as default since model optimizer default is python3 based.
sudo update-alternatives --install /usr/bin/python python /usr/bin/python3 10
Base on Ubuntu
# Environment proxy settings:
vim /etc/apt/apt.conf
Acquire::http::proxy “http://proxy.example.com:xxx/ “;
Acquire::https::proxy “http://proxy.example.com:xxx/ “;
vim /etc/environment
export http_proxy="http://proxy.example.com:xxx/"
export https_proxy=”http://proxy.example.com:xxx/”
vim ~/.bashrc
export http_proxy="http://proxy.example.com:xxx/"
export https_proxy=”http://proxy.example.com:xxx/”
# Install python3.6
mkdir /usr/local/python3
cd /usr/local/python3
wget --no-check-certificate https://www.python.org/ftp/python/3.6.5/Python-3.6.5.tgz
tar -xzvf Python-3.6.5.tgz
cd Python-3.6.5
./configure --prefix=/usr/local/python3
make
make install
# Download get-pip.py
apt-get remove python-pip python3-pip
wget https://bootstrap.pypa.io/get-pip.py
apt-get install python3-distutils
python3 get-pip.py
sudo update-alternatives --install /usr/bin/python python /usr/bin/python3 10
pip3 freeze >requirements.txt
pip3 install -r requirements.txt
# Set environment variables
Open the .bashrc file in <user_directory>:
vi <user_directory>/.bashrc
Add this line to the end of the file:
source /opt/intel/openvino/bin/setupvars.sh
Download
- · OpenVINO toolkit package: https://software.intel.com/en-us/openvino-toolkit/choosedownload/
free-download-linux
Install Guide
- · OpenVINO toolkit install guide:
https://docs.openvinotoolkit.org/latest/_docs_install_guides_installing_openvino_linux.html
# Install the Intel® Distribution of OpenVINO™ Toolkit
tar -xvzf l_openvino_toolkit_p_<version>.tgz
cd l_openvino_toolkit_p_<version>
sudo ./install.sh
# Install External software dependencies
cd /opt/intel/openvino/install_dependencies
sudo -E ./install_openvino_dependencies.sh
# Configure the Model Optimizer
cd /opt/intel/openvino/deployment_tools/model_optimizer/install_prerequisites
sudo ./install_prerequisites.sh
# Run the Verification Scripts to Verify Installation and Compile Samples
cd /opt/intel/openvino/deployment_tools/demo
./demo_squeezenet_download_convert_run.sh
./demo_security_barrier_camera.sh
BUILD Sample APP
- · Go to the directory with the build_samples.sh script and run it:
cd /opt/intel/openvino/deployment_tools/inference_engine/samples
./build_samples.sh
SSD-Mobilenet-v1
Prepare the model, datasets and related datafiles
- · Model Downloader:
cd /opt/intel/openvino/deployment_tools/tools/model_downloader/
Download all via ./downloader.py --all or download specific model via ./downloader.py –
name MODEL_NAME --output_dir LOCAL_DIR
- · Download datasets:
ImageNet 2012: http://www.image-net.org/download-images
COCO 2017: http://images.cocodataset.org/zips/val2017.zip
VOC 2007: http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
FP32 INFERENCE
Convert to IR
Execute this step then you will get frozen_inference_graph.bin/mapping/xml files under
irmodels：
python /opt/intel/openvino/deployment_tools/model_optimizer/mo.py --framework=tf --data_type=FP32 --reverse_input_channels --input_shape=[1,300,300,3] --input=image_tensor --tensorflow_use_custom_operations_config=/opt/intel/openvino/deployment_tools/model_optimizer/extensions/front/tf/ssd_v2_support.json --output=detection_scores,detection_boxes,num_detections --tensorflow_object_detection_api_pipeline_config=/home/dldt/model_downloader/public/ssd_mobilenet_v1_coco/ssd_mobilenet_v1_coco_2018_01_28/pipeline.config --input_model=/home/dldt/model_downloader/public/ssd_mobilenet_v1_coco/ssd_mobilenet_v1_coco_2018_01_28/frozen_inference_graph.pb --output_dir /home/dldt/irmodels
Convert Data Annotation Files
python /opt/intel/openvino/deployment_tools/tools/accuracy_checker_tool/convert_annotation.py mscoco_detection --annotation_file /home/dldt/datasets/annotations/instances_val2017.json --has_background True --use_full_label_map True -o /home/dldt/datasets/annotation -a mscoco_detection.pickle -m mscoco_detection.json
VALIDATE ACCURACY
Use accuracy_checker.py to verify the accuracy：
python /opt/intel/openvino/deployment_tools/tools/accuracy_checker_tool/accuracy_check.py --config /home/dldt/yml/ssd_mobilenet_v1_coco.yml -d /opt/intel/openvino/deployment_tools/tools/calibration_tool/configs/definitions.yml -M /opt/intel/openvino/deployment_tools/model_optimizer --models /home/dldt/irmodels --source /home/dldt/datasets --annotations /home/dldt/datasets/annotation
#Below are the result:
Processing info:
model: ssd_mobilenet_v1_coco
launcher: dlsdk
device: CPU
dataset: COCO2017_90cl_bkgr
OpenCV version: 4.1.2-openvino
IE version: 2.1.custom_releases/2019/R3_ac8584cb714a697a12f1f30b7a3b78a5b9ac5e05
Loaded CPU plugin version: 2.1.32974
4952 objects processed in 102.482 seconds
map: 36.09%
FP32 Inference
/root/inference_engine_samples_build/intel64/Release/benchmark_app -progress true -i /home/dldt/datasets/val2017 -b 8 -m /home/dldt/irmodels/frozen_inference_graph.xml -d CPU -api async -l /opt/intel/openvino/inference_engine/lib/intel64/libcpu_extension_avx512.so -nireq 1 -nstreams 1
#Below are the result:
[Step 11/11] Dumping statistics report
Count: 4154 iterations
Duration: 60017.08 ms
Latency: 14.39 ms
Throughput: 553.71 FPS
[Step 11/11] Dumping statistics report
Count: 4164 iterations
Duration: 60026.74 ms
Latency: 14.35 ms
Throughput: 554.95 FPS
INT8 INFERENCE
Calibration
The yml file is used to do calibration based on ssd_mobilenet_v1 model.
Below are using ssd_mobilenet_v1_coco yml file. Pls note, the file path
in the yml file need to be changed based on where you store those files.
The calibration command:
python /opt/intel/openvino/deployment_tools/tools/calibration_tool/calibrate.py --config /home/dldt/yml/ssd_mobilenet_v1_coco.yml --definition /opt/intel/openvino/deployment_tools/tools/calibration_tool/configs/definitions.yml -M /opt/intel/openvino/deployment_tools/model_optimizer --models /home/dldt/irmodels --source /home/dldt/datasets --annotations /home/dldt/datasets/annotation
After done, frozen_inference_graph_i8.bin/xml is saved in the same folder of fp32 model.
Now you need to create a yml file for int8 model.
Verify Accuracy
python /opt/intel/openvino/deployment_tools/tools/accuracy_checker_tool/accuracy_check.py --config /home/dldt/yml/ssd_mobilenet_v1_coco_i8.yml -d /opt/intel/openvino/deployment_tools/tools/calibration_tool/configs/definitions.yml -M /opt/intel/openvino/deployment_tools/model_optimizer --models /home/dldt/irmodels --source /home/dldt/datasets --annotations /home/dldt/datasets/annotation
#Below are the result:
Processing info:
model: ssd_mobilenet_v1_coco
launcher: dlsdk
device: CPU
dataset: COCO2017_90cl_bkgr
OpenCV version: 4.1.2-openvino
IE version: 2.1.custom_releases/2019/R3_ac8584cb714a697a12f1f30b7a3b78a5b9ac5e05
Loaded CPU plugin version: 2.1.32974
4952 objects processed in 97.580 seconds
map: 35.90%
INT8 Inference
/root/inference_engine_samples_build/intel64/Release/benchmark_app -progress true -i /home/dldt/datasets/val2017 -b 8 -m /home/dldt/irmodels/frozen_inference_graph_i8.xml -d CPU -api async -l /opt/intel/openvino/inference_engine/lib/intel64/libcpu_extension_avx512.so -nireq 1 -nstreams 1
#Below are the result:
[Step 11/11] Dumping statistics report
Count: 9900 iterations
Duration: 60009.23 ms
Latency: 6.04 ms
Throughput: 1319.80 FPS
[Step 11/11] Dumping statistics report
Count: 9940 iterations
Duration: 60006.40 ms
Latency: 6.01 ms
Throughput: 1325.19 FPS
Base on Docker
# Install Docker
Following https://docs.docker.com/install/linux/docker-ce/ubuntu/ to get and install docker.
# Download docker basic image
docker pull ubuntu:16.04
# Start the container
docker run -itd --privileged ubuntu:16.04 bash
docker container ls # Get the container id, $container_id=CONTAINER ID
docker exec -it $container_id bash #enter container
# Environment proxy settings:
export http_proxy="http://proxy.example.com:xxx/"
export https_proxy=”http://proxy.example.com:xxx/”
#Install dependable environment
apt-get update
apt-get install curl lsb-release
apt install apt-utils
apt-get install vim
apt-get install wget
apt-get install sudo
apt-get install build-essential
apt-get install zlib1g-dev
apt-get install cpio
apt-get install pciutils
apt-get install numactl
apt-get install libnuma-dev
apt-get install libgtk-3-0
# Then next steps follow “Base on Ubuntu”
浙公网安备 33010602011771号