darknet模型转openvino

0. 拢共分为两步，一是darknet转tensorflow，二是tensorflow转openvino

1. darknet转tensorflow

• conda创建python3.6虚拟环境
• 安装tensorflow==1.11.0
• git clone https://github.com/mystic123/tensorflow-yolo-v3.git
• cd tensorflow-yolo-v3
• python convert_weights_pb.py --class_names coco.names --data_format NHWC --weights_file yolov3.weights
• 就会在当前文件夹生成frozen_darknet_yolov3_model.pb

2. tensorflow转openvino

• cd /opt/intel/openvino/deployment_tools/model_optimizer
• pip install -r requirements_tf.txt
• python3 mo_tf.py --input_model /home/lwd/code/dl/tensorflow-yolo-v3/frozen_darknet_yolov3_model.pb --tensorflow_use_custom_operations_config extensions/front/tf/yolo_v3.json --batch 1 -o /home/lwd/code/dl/ie/model

3. 测试

cd /opt/intel/openvino/deployment_tools/open_model_zoo/demos/python_demos/object_detection_demo_yolov3_async
python object_detection_demo_yolov3_async.py -m /opt/intel/openvino/deployment_tools/model_optimizer/frozen_darknet_yolov3_model.xml -i /home/lwd/Videos/VID_20200615_103705.mp4 -d CPU