使用OpenVINO转换ssd_resnet_50_fpn_co
ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03模型下载地址:TensorFlow 1 Detection Model Zoo
OpenVINO版本:2021.3.394
第一步:新建tf1_ov虚拟环境,注意Python=3.6。选择3.7或3.8会导致install_prerequisites_tf.bat安装tensorflow版本出错。在pip库中,tensorflow 1.15支持的Python版本最高为3.6
第二步:运行install_prerequisites_tf.bat,配置model optimizer的运行环境
第三步,执行命令:python mo_tf.py --input_model d:\ov_mo\ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03\frozen_inference_graph.pb --tensorflow_object_detection_api_pipeline_config d:\ov_mo\ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03\pipeline.config --output_dir d:\ov_mo\ssd_resnet50_v1_fpn_shared_box_predictor_640x640_coco14_sync_2018_07_03\ --model_name ssd_resnet50_tf1_ov --data_type FP16 --reverse_input_channels --transformations_config extensions\front\tf\ssd_v2_support.json
ssd_resnet50_v1_fpn.png参考文献:model_optimizer_args