topic TF2 [Object Detection API] Converting mobilenet-SSD models into .tflite uint8 format in eIQ Machine Learning Software

topic TF2 [Object Detection API] Converting mobilenet-SSD models into .tflite uint8 format in eIQ Machine Learning Software https://community.nxp.com/t5/eIQ-Machine-Learning-Software/TF2-Object-Detection-API-Converting-mobilenet-SSD-models-into/m-p/1345027#M486 <P><STRONG>1. Prepare the environment</STRONG></P> <LI-CODE lang="markup">pip install tensorflow==2.5.0</LI-CODE> <P><STRONG>2. Install  tf2 Object detect API</STRONG></P> <P><SPAN>Make sure you have </SPAN><A href="https://grpc.io/docs/protoc-installation/#install-using-a-package-manager" rel="nofollow" target="_blank">protobuf compiler</A><SPAN> version >= 3.0, by typing </SPAN><CODE>protoc --version</CODE><SPAN>, or install it on Ubuntu by typing </SPAN><CODE>apt install protobuf-compiler</CODE><SPAN>.</SPAN></P> <LI-CODE lang="markup">git clone https://github.com/tensorflow/models.git # remember to activate your python environment first cd models/research # compile protos: protoc object_detection/protos/*.proto --python_out=. # Install TensorFlow Object Detection API as a python package: cp object_detection/packages/tf2/setup.py . python -m pip install .</LI-CODE> <P><STRONG>3. export tflite graph and convert to quante tflite module.</STRONG></P> <LI-CODE lang="markup">1) $cd models $wget http://download.tensorflow.org/models/object_detection/tf2/20200711/ ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz 2)tar -zxvf ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz 3)tar -zxvf convert.tar.gz</LI-CODE> <P>3.1   modify export_tflite.sh</P> <LI-CODE lang="markup">export_tflite.sh 1 model_dir=../ssd_mobilenet_v2_320x320_coco17_tpu-8 <--pre-trained model path 2 out_dir=$model_dir/exported_tflite 3 mkdir -p $out_dir </LI-CODE> <P>3.2  export tflite graph</P> <LI-CODE lang="markup">./export_tflite.sh</LI-CODE> <P>Use <STRONG>EIQ toolkit </STRONG> model tool to open the saved model, the input tensor shape is 300x300</P> <P><span class="lia-inline-image-display-wrapper lia-image-align-inline" image-alt="gnar_fang_0-1632395542817.png" style="width: 400px;"><img src="https://community.nxp.com/t5/image/serverpage/image-id/157071i52CFAFAEA422D3E1/image-size/medium?v=v2&px=400" role="button" title="gnar_fang_0-1632395542817.png" alt="gnar_fang_0-1632395542817.png" /></span></P> <P>3.3  modify convert.py</P> <DIV> <DIV><SPAN><STRONG>line 3 TEST_DIR </STRONG>      The parent directory of the sample images</SPAN></DIV> <DIV><SPAN><STRONG>line 4  IMAGE_SIZE  </STRONG>The model input tensor shape</SPAN></DIV> <DIV><SPAN><STRONG>line 8</STRONG> quante              The sample images folder</SPAN></DIV> <DIV> </DIV> <DIV>3.4  run converter script.</DIV> <DIV><LI-CODE lang="markup">python convert.py</LI-CODE></DIV> </DIV> <P> </P> <P>4. test performance on imx8mpevk</P> <LI-CODE lang="markup">root@imx8mpevk:/usr/bin/tensorflow-lite-2.5.0/examples# ./label_image -m ~/ssd_mobilenet_v2_quant.tflite -a 1 INFO: Applied NNAPI delegate. INFO: invoked INFO: average time: 27.365 ms </LI-CODE> <P> </P> <P> </P> Thu, 23 Sep 2021 11:29:04 GMT gnar_fang 2021-09-23T11:29:04Z TF2 [Object Detection API] Converting mobilenet-SSD models into .tflite uint8 format https://community.nxp.com/t5/eIQ-Machine-Learning-Software/TF2-Object-Detection-API-Converting-mobilenet-SSD-models-into/m-p/1345027#M486 <P>This is a guide for Post trained quantization for SSD models.</P> <P>Take <A href="http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz" rel="nofollow">SSD MobileNet v2 320x320</A> as an example.</P> <P> </P> <P>Environment</P> <P>TensorFlow == 2.5</P> <P> </P> Thu, 23 Sep 2021 11:29:04 GMT https://community.nxp.com/t5/eIQ-Machine-Learning-Software/TF2-Object-Detection-API-Converting-mobilenet-SSD-models-into/m-p/1345027#M486 gnar_fang 2021-09-23T11:29:04Z