{"id":3032,"date":"2020-03-24T19:28:26","date_gmt":"2020-03-24T19:28:26","guid":{"rendered":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/"},"modified":"2023-03-03T19:45:32","modified_gmt":"2023-03-03T19:45:32","slug":"automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning","status":"publish","type":"post","link":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/","title":{"rendered":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-post\" data-elementor-id=\"3032\" class=\"elementor elementor-3032\" data-elementor-post-type=\"post\">\n\t\t\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-2ac04a2 elementor-section-boxed elementor-section-height-default elementor-section-height-default qodef-elementor-content-no\" data-id=\"2ac04a2\" data-element_type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-4a29edab\" data-id=\"4a29edab\" data-element_type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t<div class=\"elementor-element elementor-element-e62a55d elementor-widget elementor-widget-text-editor\" data-id=\"e62a55d\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t<style>\/*! elementor - v3.21.0 - 25-04-2024 *\/\n.elementor-widget-text-editor.elementor-drop-cap-view-stacked .elementor-drop-cap{background-color:#69727d;color:#fff}.elementor-widget-text-editor.elementor-drop-cap-view-framed .elementor-drop-cap{color:#69727d;border:3px solid;background-color:transparent}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap{margin-top:8px}.elementor-widget-text-editor:not(.elementor-drop-cap-view-default) .elementor-drop-cap-letter{width:1em;height:1em}.elementor-widget-text-editor .elementor-drop-cap{float:left;text-align:center;line-height:1;font-size:50px}.elementor-widget-text-editor .elementor-drop-cap-letter{display:inline-block}<\/style>\t\t\t\t<p>Machine learning (ML) with neural networks is enabling exciting new inference capabilities for software. Typically, ML models have run in the cloud, which meant that, to make a classification or prediction, you needed to send the text, sound, or images over the network to an external vendor. The performance of such solutions strongly depends on the network bandwidth and latency. Moreover, sending your data to an external service introduces potential privacy concerns.<\/p><p>To solve those problems, AI is now moving to edge devices. Arm is leading this revolution by providing accelerated hardware IPs and middle-layer solutions.<\/p><p>Arm NN, Arm\u2019s open-source software tool, helps developers build an application once and port it to different platforms seamlessly and effortlessly.<\/p><p>Arm NN and the Arm Compute Library are open-source libraries for optimizing ML on Arm-based processors and IoT edge devices. ML workloads can run entirely on edge devices, allowing sophisticated AI-enabled software to run almost anywhere, even without network access.<\/p><p>In practice, this means you can train your models using the most popular ML frameworks. You can then load models and optimize them on your edge device to perform fast inference, eliminating network and privacy problems.<\/p><p>To demonstrate ML with edge computing resources, we&#8217;ll use Arm NN APIs to classify images of trash from a webcam attached to a Raspberry Pi. The Raspberry Pi will display classification results.<\/p><h2>What do we need for this tutorial?<\/h2><p>Here&#8217;s what we used to create the device and application for this tutorial:<\/p><ul style=\"font-size: 16px; background-color: #ffffff;\"><li style=\"font-size: 16px;\">A Raspberry Pi device. The tutorial has been verified with Pi 2, Pi 3, and Pi 4 Model B.<\/li><li style=\"font-size: 16px;\">A MicroSD card<\/li><li style=\"font-size: 16px;\">A USB or MIPI camera module for the Raspberry Pi<\/li><li style=\"font-size: 16px;\">If you want to build your own Arm NN library, you also need a Linux host machine or a computer with Linux virtual environment installed.<\/li><li style=\"font-size: 16px;\">Glass, paper, cardboard, plastic, metal or any trash that your Raspberry Pi can help you classify!<\/li><\/ul><p><img decoding=\"async\" src=\"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/image001-5.png\" width=\"453px\" height=\"281px\" \/><\/p><h2>Device Setup<\/h2><p>I used the Raspberry Pi 4 Model B with a quad-core ARM Cortex A72 CPU, 1GB onboard memory, an 8GB MicroSD card, a\u00a0<a href=\"https:\/\/www.raspberrypi.org\/products\/raspberry-pi-usb-wifi-dongle\/\">WiFi dongle<\/a>, a USB camera (Microsoft HD-3000).<\/p><p>Before powering up the device, you need to install Raspbian on the MicroSD card as described on the\u00a0<a href=\"https:\/\/projects.raspberrypi.org\/en\/projects\/raspberry-pi-setting-up\/3\">Setting up your Raspberry Pi tutorial<\/a>. I used NOOBS for easy setup.<\/p><p>I then booted my Raspberry Pi, configured Raspbian, and set up virtual network computing (VNC) for remote access to my Raspberry Pi. You\u2019ll find detailed instructions for setting up the VNC at the RaspberryPi.org\u00a0<a href=\"https:\/\/www.raspberrypi.org\/documentation\/remote-access\/vnc\/README.md\">VNC (Virtual Network Computing)<\/a>\u00a0page.<\/p><p>After configuring the hardware, I started on the software. The solution is comprised of three components:<\/p><ul style=\"font-size: 16px; background-color: #ffffff;\"><li style=\"font-size: 16px;\"><em>camera.hpp<\/em>\u00a0implements helper methods for capturing images from a webcam.<\/li><li style=\"font-size: 16px;\"><em>ml.hpp<\/em>\u00a0contains methods to load the ML model and classify trash images based on camera input.<\/li><li style=\"font-size: 16px;\"><em>main.cpp<\/em>\u00a0contains the main method, which wires the above components together. This is the entry point of the application.<\/li><\/ul><p>All of these components are discussed below. Everything you see here was created with the Geany editor, which comes with the default Raspbian installation.<\/p><h2>Setting up the Camera<\/h2><p>To get images from the webcam, I used OpenCV, the computer vision (CV) open-source library. This library provides a convenient interface to capture and process images. You can easily use the same API across different applications and devices, ranging from IoT to mobile to desktop.<\/p><p>The easiest way to incorporate OpenCV into your IoT Raspbian apps is to install the libopencv-dev package through apt-get:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-47cc97d elementor-widget elementor-widget-code-highlight\" data-id=\"47cc97d\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-javascript line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-javascript\">\n\t\t\t\t\t<xmp>sudo apt-get update\r\n\r\nsudo apt-get upgrade\r\nsudo apt-get install libopencv-dev<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-ce689ba elementor-widget elementor-widget-text-editor\" data-id=\"ce689ba\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>After the packages are downloaded and installed, you\u2019re ready to capture images from a webcam. I started by implementing two methods:\u00a0<code style=\"font-size: 16px;\">grabFrame<\/code>\u00a0and\u00a0<code style=\"font-size: 16px;\">showFrame<\/code>\u00a0(see\u00a0<em>camera.hpp<\/em>\u00a0in the companion code):<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-f2ea6bf elementor-widget elementor-widget-code-highlight\" data-id=\"f2ea6bf\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>Mat grabFrame()\r\n{\r\n  \/\/ Open default camera\r\n  VideoCapture cap(0);\r\n\r\n  \/\/ If camera was open, get the frame\r\n  Mat frame;\r\n  if (cap.isOpened())\r\n  {\r\ncap >> frame;\r\n    imwrite(\"image.jpg\", frame); \r\n  }\r\n  else\r\n  {\r\n    printf(\"No valid cameran\");\r\n  }\r\n\r\n  return frame;\r\n}\r\n \r\nvoid showFrame(Mat frame)\r\n{\r\n  if (!frame.empty())\r\n  {\r\n    imshow(\"Image\", frame);\r\n    waitKey(0);\r\n  }\r\n}\r\n<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-97a0238 elementor-widget elementor-widget-text-editor\" data-id=\"97a0238\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<pre style=\"font-size: 16px;\"><span style=\"color: var( --e-global-color-text ); font-family: var( --e-global-typography-text-font-family ), Sans-serif; font-weight: var( --e-global-typography-text-font-weight ); font-size: 1rem;\">The first method, <\/span><code style=\"font-size: 1rem; color: var( --e-global-color-text ); font-weight: var( --e-global-typography-text-font-weight );\">grabFrame<\/code><span style=\"color: var( --e-global-color-text ); font-family: var( --e-global-typography-text-font-family ), Sans-serif; font-weight: var( --e-global-typography-text-font-weight ); font-size: 1rem;\">, opens the default web camera (index of 0) and captures a single frame. Note that the C++ OpenCV interface uses the Mat class to represent images, so grabFrame returns objects of this type. The raw image data can be accessed by reading the data member of the Mat class.<\/span><\/pre><p>The second method,\u00a0<code>showFrame<\/code>, is used to display the captured image. To that end,\u00a0<code>showFrame<\/code>\u00a0uses OpenCV\u2019s imshow method to create a window in which the image will be displayed. I then invoke the\u00a0<code>waitKey<\/code>\u00a0method, which forces the image window to be displayed until the user hits a key on the keyboard. I used the\u00a0<code>waitKey<\/code>\u00a0parameter to specify the timeout. Here, I used an infinite timeout, represented by 0.<\/p><p>To test the above methods, I invoked them within the main method (<em>main.cpp<\/em>):<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-99811ec elementor-widget elementor-widget-code-highlight\" data-id=\"99811ec\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>int main()\r\n{\r\n  \/\/ Grab image\r\n  Mat frame = grabFrame();\r\n \r\n  \/\/ Display image\r\n  showFrame(frame);\r\n \r\n  return 0;\r\n}<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-6b781cc elementor-widget elementor-widget-text-editor\" data-id=\"6b781cc\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>To build the app, I used the g++ command and linked the OpenCV libraries through pkg-config:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-079e0c2 elementor-widget elementor-widget-code-highlight\" data-id=\"079e0c2\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>g++ main.cpp -o trashClassifier 'pkg-config --cflags --libs opencv'<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-a51c8cb elementor-widget elementor-widget-text-editor\" data-id=\"a51c8cb\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Afterward, I executed the app to capture a single image:<\/p><p><img decoding=\"async\" src=\"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/image002-6.png\" width=\"417px\" height=\"333px\" \/><\/p><h2>Trash Dataset and Model Training<\/h2><p>I trained the TensorFlow classification model using an existing dataset created by Gary Thung and available from his\u00a0<a href=\"https:\/\/github.com\/garythung\/trashnet\">trashnet Github repo<\/a>.<\/p><p>To train the model, I followed the steps of the\u00a0<a href=\"https:\/\/www.tensorflow.org\/tutorials\/images\/classification\">TensorFlow image classification tutorial<\/a>. However, I trained the model on images of 256&#215;192 pixels \u2014 half the width and height of the original images from the trash dataset. Here is what our model consists of:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-e56f4bc elementor-widget elementor-widget-code-highlight\" data-id=\"e56f4bc\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp># Building the model\r\nmodel = Sequential()\r\n# 3 convolutional layers\r\nmodel.add(Conv2D(32, (3, 3), input_shape = (IMG_HEIGHT, IMG_WIDTH, 3)))\r\nmodel.add(Activation(\"relu\"))\r\nmodel.add(MaxPooling2D(pool_size=(2,2)))\r\n\r\nmodel.add(Conv2D(64, (3, 3)))\r\nmodel.add(Activation(\"relu\"))\r\nmodel.add(MaxPooling2D(pool_size=(2,2)))\r\n\r\nmodel.add(Conv2D(64, (3, 3)))\r\nmodel.add(Activation(\"relu\"))\r\nmodel.add(MaxPooling2D(pool_size=(2,2)))\r\nmodel.add(Dropout(0.25))\r\n\r\n# 2 hidden layers\r\nmodel.add(Flatten())\r\nmodel.add(Dense(128))\r\nmodel.add(Activation(\"relu\"))\r\n\r\nmodel.add(Dense(128))\r\nmodel.add(Activation(\"relu\"))\r\n\r\n# The output layer with 6 neurons, for 6 classes\r\nmodel.add(Dense(6))\r\nmodel.add(Activation(\"softmax\"))<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-ff35e10 elementor-widget elementor-widget-text-editor\" data-id=\"ff35e10\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>The model achieved approximately 83% accuracy. We use\u00a0<code style=\"font-size: 16px;\">tf.lite.TFLiteConverter<\/code>\u00a0to convert it to TensorFlow Lite format\u00a0<code style=\"font-size: 16px;\">trash_model.tflite<\/code>.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-67d1407 elementor-widget elementor-widget-code-highlight\" data-id=\"67d1407\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>converter = tf.lite.TFLiteConverter.from_keras_model_file('model.h5') \r\nmodel = converter.convert()\r\nfile = open('model.tflite' , 'wb') \r\nfile.write(model)<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-ccdcc3e elementor-widget elementor-widget-text-editor\" data-id=\"ccdcc3e\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<h2>Setting up the Arm NN SDK<\/h2><p>The next step is to prepare the Arm NN SDK. To generate the Arm NN library for your\u00a0<a href=\"https:\/\/contentlab.com\/dog-mood-detector-with-google-cloud-and-arm-based-raspberry-pi\/\">Raspberry Pi<\/a>, you can either follow the Arm\u00a0<a href=\"https:\/\/developer.arm.com\/solutions\/machine-learning-on-arm\/developer-material\/how-to-guides\/cross-compiling-arm-nn-for-the-raspberry-pi-and-tensorflow\">Cross-compile Arm NN and Tensorflow for the Raspberry Pi<\/a>\u00a0tutorial or run the automated script from the Arm\u00a0<a href=\"https:\/\/github.com\/ARM-software\/Tool-Solutions\/tree\/master\/ml-tool-examples\/build-armnn\">Tool-Solutions Github repository<\/a>\u00a0to cross-compile the SDK. The Arm NN 19.08 binary tar file for Raspberry Pi can be found\u00a0<a href=\"https:\/\/github.com\/ARM-software\/armnn\/releases\">on GitHub<\/a>.<\/p><p>No matter which approach you choose, copy the resulting tar (armnn-dist) to your Raspberry Pi. In this case, I use the VNC to transfer files between my development PC and the Raspberry Pi.<\/p><p>Next, set the\u00a0<code>LD_LIBRARY_PATH<\/code>\u00a0environment variable to point to the\u00a0<em>armnn\/lib<\/em>\u00a0subfolder of\u00a0<code>armnn-dist<\/code>:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-bebfda0 elementor-widget elementor-widget-code-highlight\" data-id=\"bebfda0\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>export LD_LIBRARY_PATH=\/home\/pi\/armnn-dist\/armnn\/lib<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7297e6c elementor-widget elementor-widget-text-editor\" data-id=\"7297e6c\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Here, I assume that the\u00a0<code>armnn-dist<\/code>\u00a0is located under the\u00a0<em>home\/pi<\/em>\u00a0folder.<\/p><h2>Use Arm NN for on-device ML inference<\/h2><h3>Load the model output labels<\/h3><p>You must use the model output labels to interpret the outputs of the model. In our code, we create a string vector to store the 6 classes.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-c695cf1 elementor-widget elementor-widget-code-highlight\" data-id=\"c695cf1\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>const std::vector<std::string> modelOutputLabels = {\"cardboard\", \"glass\", \"metal\", \"paper\", \"plastic\", \"trash\"};<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7b551fc elementor-widget elementor-widget-text-editor\" data-id=\"7b551fc\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<h3>Load and pre-process an input image<\/h3><p>We must pre-process images before the model can use them as inputs. The pre-processing method that we use depends on the framework, model, or model data type.<\/p><p>Our model has an input of a Conversion 2D layer, with id \u2018conv2d_input\u2019. Its output is a softmax activation layer, with id \u2018<code>activation_5\/Softmax<\/code>\u2019. The model properties are extracted by using Tensorboard, a visualization tool provided by TensorFlow for model inspection.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-9e8fe37 elementor-widget elementor-widget-code-highlight\" data-id=\"9e8fe37\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>const std::string inputName = \"conv2d_input\";\r\nconst std::string outputName = \"activation_5\/Softmax\";\r\n\r\nconst unsigned int inputTensorWidth = 256;\r\nconst unsigned int inputTensorHeight = 192;\r\nconst unsigned int inputTensorBatchSize = 32;\r\nconst armnn::DataLayout inputTensorDataLayout = armnn::DataLayout::NHWC;<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-a086023 elementor-widget elementor-widget-text-editor\" data-id=\"a086023\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\tNote that the batch size used for training is equal to 32, so you need to provide at least 32 images for training and validation.\n\nThe following code loads and pre-processes an image the camera captures:\n\n<pre>\/\/ Load and preprocess input image\nconst std::vector&lt;tcontainer&gt; inputDataContainers =\n{ PrepareImageTensor&lt;uint8_t&gt;(\"image.jpg\" ,\n inputTensorWidth, inputTensorHeight,\n normParams,\n inputTensorBatchSize,\n inputTensorDataLayout) } ;<\/pre>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-34b2569 elementor-widget elementor-widget-code-highlight\" data-id=\"34b2569\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language- line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-\">\n\t\t\t\t\t<xmp>\/\/ Load and preprocess input image\nconst std::vector<tcontainer> inputDataContainers =\n{ PrepareImageTensor<uint8_t>(\"image.jpg\" ,\n inputTensorWidth, inputTensorHeight,\n normParams,\n inputTensorBatchSize,\n inputTensorDataLayout) } ;<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-3742952 elementor-widget elementor-widget-text-editor\" data-id=\"3742952\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>All logic related to loading the ML model and making predictions is contained in the ml.hpp file,<\/p><h3>Create a parser and load the network<\/h3><p>The next step when working with Arm NN is to create a parser object that will be used to load your network file. Arm NN has parsers for a variety of model file types, including TFLite, ONNX, Caffe etc. Parsers handle creation of the underlying Arm\u00a0<a href=\"https:\/\/contentlab.com\/c-neural-network-in-a-weekend\/\">NN<\/a>\u00a0graph so you don&#8217;t need to construct your model graph by hand.<\/p><p>In this example, we are using TFLite model so we create the TfLite parser to load the model from the specified path.<\/p><p>The most important method from\u00a0<em>ml.hpp<\/em>\u00a0is\u00a0<code>loadModelAndPredict<\/code>. It first creates the TensorFlow model parser:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-7cff0ee elementor-widget elementor-widget-code-highlight\" data-id=\"7cff0ee\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Import the TensorFlow model. \r\n\/\/ Note: use CreateNetworkFromBinaryFile for .tflite files.\r\narmnnTfLiteParser::ITfLiteParserPtr parser = \r\n  armnnTfLiteParser::ITfLiteParser::Create();\r\narmnn::INetworkPtr network = \r\n  parser->CreateNetworkFromBinaryFile(\"trash_model.tflite\");<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-faecf89 elementor-widget elementor-widget-text-editor\" data-id=\"faecf89\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Next, I invoke the a<code>rmnnTfLiteParser::ITfLiteParser::Create<\/code>\u00a0method and use this parser to load trash_model.tflite.<\/p><p>Once the model is parsed, you create the bindings to those layers using the\u00a0<code>GetNetworkInputBindingInfo<\/code>\u00a0\/\u00a0<code>GetNetworkOutputBindingInfo<\/code>\u00a0method:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-ed1f130 elementor-widget elementor-widget-code-highlight\" data-id=\"ed1f130\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Find the binding points for the input and output nodes\r\nconst size_t subgraphId = 0;\r\narmnnTfParser::BindingPointInfo inputBindingInfo = \r\n    parser->GetNetworkInputBindingInfo(subgraphId, inputName);\r\narmnnTfParser::BindingPointInfo outputBindingInfo = \r\n    parser->GetNetworkOutputBindingInfo(subgraphId, outputName);<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-a55a3e0 elementor-widget elementor-widget-text-editor\" data-id=\"a55a3e0\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>You must prepare a container to receive the output of the model. The output tensor size is equal to the number of model output labels. This is implemented as below:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-fa57ecd elementor-widget elementor-widget-code-highlight\" data-id=\"fa57ecd\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Output tensor size is equal to the number of model output labels\r\nconst unsigned int outputNumElements = modelOutputLabels.size();\r\nstd::vector<TContainer> outputDataContainers = { std::vector<uint8_t>(outputNumElements)};<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-19f4fbb elementor-widget elementor-widget-text-editor\" data-id=\"19f4fbb\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<h3>Choose backends, create runtime, and optimize the model<\/h3><p>You must optimize your network and load it onto a compute device. The Arm NN SDK supports optimized execution backends on Arm CPU, Mali GPU, and DSP devices. Backends are identified by a string that must be unique across backends. You can specify one or more backend in order of preference.<\/p><p>In our code, Arm NN decides which layers are supported by the backend. First the CPU is checked. If one or more of the layers can&#8217;t be run on CPU, it will fall back first to the reference implementation.<\/p><p>After specifying the backend list, you can create a runtime and optimize the network in the runtime context. The backends may choose to implement backend-specific optimizations. Arm NN splits the graph into subgraphs based on backends, it calls an optimize subgraph function on each of them, and substitutes the corresponding sub-graph in the original graph with its optimized version when possible.<\/p><p>Once this is done,\u00a0<code>LoadNetwork<\/code>\u00a0creates the backend-specific workloads for the layers, it creates a backend specific workload factory and calls this to create the workloads. The input image is wrapped in a const tensor and bound to the input tensor.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-02036a1 elementor-widget elementor-widget-code-highlight\" data-id=\"02036a1\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Optimize the network for a specific runtime compute \r\n\/\/ device, e.g. CpuAcc, GpuAcc\r\narmnn::IRuntime::CreationOptions options;\r\narmnn::IRuntimePtr runtime = armnn::IRuntime::Create(options);\r\narmnn::IOptimizedNetworkPtr optNet = armnn::Optimize(*network,\r\n   {armnn::Compute::CpuAcc, armnn::Compute::CpuRef},  \r\n   runtime->GetDeviceSpec());<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-bb25205 elementor-widget elementor-widget-text-editor\" data-id=\"bb25205\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>The Arm NN SDK inference engine provides a bridge between existing neural network frameworks and Arm Cortex-A CPUs, Arm Mali GPUs, and DSPs. Running ML inference with Arm NN SDK ensures the ML algorithms will be optimized for the underlying hardware.<\/p><p>After optimization, you load the network to the runtime:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-537eb1e elementor-widget elementor-widget-code-highlight\" data-id=\"537eb1e\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Load the optimized network onto the runtime device\r\narmnn::NetworkId networkIdentifier;\r\nruntime->LoadNetwork(networkIdentifier, std::move(optNet));<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-1039a79 elementor-widget elementor-widget-text-editor\" data-id=\"1039a79\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Then, perform predictions using the\u00a0<code style=\"font-size: 16px;\">EnqueueWorkload<\/code>\u00a0method:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-9bdb2e3 elementor-widget elementor-widget-code-highlight\" data-id=\"9bdb2e3\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>\/\/ Predict\r\narmnn::Status ret = runtime->EnqueueWorkload(networkIdentifier,\r\n      armnnUtils::MakeInputTensors(inputBindings, inputDataContainers),\r\n      armnnUtils::MakeOutputTensors(outputBindings, outputDataContainers));<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-145a64b elementor-widget elementor-widget-text-editor\" data-id=\"145a64b\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>The last step gives you the prediction result.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-9015237 elementor-widget elementor-widget-code-highlight\" data-id=\"9015237\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>std::vector<uint8_t> output = boost::get<std::vector<uint8_t>>(outputDataContainers[0]);\r\n\r\nsize_t labelInd = std::distance(output.begin(), std::max_element(output.begin(), output.end()));\r\nstd::cout << \"Prediction: \";\r\nstd::cout << modelOutputLabels[labelInd] << std::endl;<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-bb728de elementor-widget elementor-widget-text-editor\" data-id=\"bb728de\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>In the above example, the only aspect that\u2019s inherently related to the ML framework is the part where you load the model and configure bindings. Everything else is independent of the ML framework. Hence, you can easily switch among various ML models without modifying other parts of your application.<\/p><h2>Putting Things Together and Building the App<\/h2><p>Finally, I put all of the components together within the main method (<em>main.cpp<\/em>):<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-65c0c21 elementor-widget elementor-widget-code-highlight\" data-id=\"65c0c21\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>#include \"camera.hpp\"\r\n#include \"ml.hpp\"\r\n \r\nint main()\r\n{\r\n  \/\/ Grab frame from the camera\r\n  grabFrame(true);\r\n \r\n  \/\/ Load ML model and predict\r\n  loadModelAndPredict();\r\n \r\n  return 0;\r\n}<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-19fb109 elementor-widget elementor-widget-text-editor\" data-id=\"19fb109\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Note that grabFrame has an extra parameter. When this parameter is true, the camera image is converted to grayscale and resized to 256&#215;192 pixels to match the input format of the ML model, and the converted image is then passed to the loadModelAndPredict method.<\/p><p>To build the app, use the g++ command and link OpenCV and the Arm NN SDK:<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-b4b420e elementor-widget elementor-widget-code-highlight\" data-id=\"b4b420e\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-cpp line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-cpp\">\n\t\t\t\t\t<xmp>g++ main.cpp -o trashClassifier 'pkg-config --cflags --libs opencv' -I\/home\/pi\/armnn-dist\/armnn\/include -I\/home\/pi\/armnn-dist\/boost\/include -L\/home\/pi\/armnn-dist\/armnn\/lib -larmnn -lpthread -linferenceTest -lboost_system -lboost_filesystem -lboost_program_options -larmnnTfLiteParser -lprotobuf<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-35747ae elementor-widget elementor-widget-text-editor\" data-id=\"35747ae\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p>Again, I assume that Arm NN SDK is in the\u00a0<em>home\/pi\/armnn-dist<\/em>\u00a0folder. Launch the application and take a photo of some cardboard.<\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-46d6834 elementor-widget elementor-widget-code-highlight\" data-id=\"46d6834\" data-element_type=\"widget\" data-widget_type=\"code-highlight.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<div class=\"prismjs-default copy-to-clipboard \">\n\t\t\t<pre data-line=\"\" class=\"highlight-height language-javascript line-numbers\">\n\t\t\t\t<code readonly=\"true\" class=\"language-javascript\">\n\t\t\t\t\t<xmp>pi@raspberrypi:~\/ $ .\/trashClassifier\r\nArmNN v20190800\r\n\r\nRunning network...\r\nPrediction: cardboard<\/xmp>\n\t\t\t\t<\/code>\n\t\t\t<\/pre>\n\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-6be7c5ff elementor-widget elementor-widget-text-editor\" data-id=\"6be7c5ff\" data-element_type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t<p><\/p>\n<pre><span style=\"color: var( --e-global-color-text ); font-family: var( --e-global-typography-text-font-family ), Sans-serif; font-weight: var( --e-global-typography-text-font-weight ); font-size: 1rem;\">If, during app execution, you see the message, \"error while loading shared libraries: libarmnn.so: cannot open shared object file: No such file or directory,\" check to make sure your <\/span><code style=\"color: var( --e-global-color-text ); font-weight: var( --e-global-typography-text-font-weight ); font-size: 1rem;\">LD_LIBRARY_PATH<\/code><span style=\"color: var( --e-global-color-text ); font-family: var( --e-global-typography-text-font-family ), Sans-serif; font-weight: var( --e-global-typography-text-font-weight ); font-size: 1rem;\"> is set correctly.<\/span><br><\/pre>\n<h2>Wrapping Up<\/h2>\n<p>You could also improve the app by triggering image capture and recognition via an external signal. To do this, you\u2019d modify the <code>loadAndPredict<\/code> method from the <em>ml.hpp<\/em> module. You need to separate model loading from the prediction (inference).<\/p>\n<p>This article shows how you can move AI from the cloud to the edge. Arm leads in this transition with hardware that\u2019s tailored for inference and software components to help you implement your solutions no matter which ML framework you choose: TensorFlow, TensorFlow Core, Caffe, or any other ML framework.<\/p>\n<p>You can also refer to Arm NN how to guides for additional details:<\/p>\n<ul>\n<li><a href=\"https:\/\/developer.arm.com\/solutions\/machine-learning-on-arm\/developer-material\/how-to-guides\/configuring-the-arm-nn-sdk-build-environment-for-tensorflow-lite\">Confirm Arm NN SDK build environment for TensorFlow Lite<\/a><\/li>\n<li><a href=\"https:\/\/developer.arm.com\/solutions\/machine-learning-on-arm\/developer-material\/how-to-guides\/deploying-a-quantized-tensorflow-lite-mobilenet-v1-model-using-the-arm-nn-sdk\">Deploy a quantized TensorFlow Lite MobileNet v1 model using Arm NN<\/a><br><br><\/li>\n<\/ul>\n<h4>About Arm<\/h4>\n<p>Check out the Arm solutions <a href=\"https:\/\/www.arm.com\/solutions\/artificial-intelligence\">page<\/a> to learn more about what we are doing in AI and ML. If you are a business that wants to partner with Arm, check out our <a href=\"http:\/\/arm.com\/AI-Partner-Program\">Arm AI Partner Program<\/a>.<\/p>\n<p><em><span class=\"TextRun SCXW156001673 BCX2\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW156001673 BCX2\">If you\u2019re interested in developing expert technical content that performs, <\/span><\/span><a class=\"Hyperlink SCXW156001673 BCX2\" href=\"https:\/\/contentlab.com\/contact\/\" target=\"_blank\" rel=\"noreferrer noopener\"><span class=\"TextRun Underlined SCXW156001673 BCX2\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW156001673 BCX2\" data-ccp-charstyle=\"Hyperlink\">let\u2019s have a conversation<\/span><\/span><\/a><span class=\"TextRun SCXW156001673 BCX2\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW156001673 BCX2\"> today.<\/span><\/span><\/em><\/p>\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>In this article we&#8217;ll use Arm NN APIs to classify images of trash from a webcam attached to a Raspberry Pi.<\/p>\n","protected":false},"author":3,"featured_media":5331,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[322,278,18],"tags":[311,254,312],"hashtags":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v22.5 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab\" \/>\n<meta property=\"og:description\" content=\"In this article we&#039;ll use Arm NN APIs to classify images of trash from a webcam attached to a Raspberry Pi.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\" \/>\n<meta property=\"og:site_name\" content=\"Technical Content Marketing | ContentLab\" \/>\n<meta property=\"article:published_time\" content=\"2020-03-24T19:28:26+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-03-03T19:45:32+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1746\" \/>\n\t<meta property=\"og:image:height\" content=\"530\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"ContentLab\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@ContentLabIO\" \/>\n<meta name=\"twitter:site\" content=\"@ContentLabIO\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"ContentLab\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\"},\"author\":{\"name\":\"ContentLab\",\"@id\":\"https:\/\/contentlab.com\/#\/schema\/person\/e1b8cacc1019e83e8d6cd9ab1a049532\"},\"headline\":\"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning\",\"datePublished\":\"2020-03-24T19:28:26+00:00\",\"dateModified\":\"2023-03-03T19:45:32+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\"},\"wordCount\":1864,\"publisher\":{\"@id\":\"https:\/\/contentlab.com\/#organization\"},\"image\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png\",\"keywords\":[\"Machine Learning\",\"Neural Networks\",\"Raspberry Pi\"],\"articleSection\":[\"AI\",\"Hands-on Tutorial\",\"Portfolio\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\",\"url\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\",\"name\":\"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab\",\"isPartOf\":{\"@id\":\"https:\/\/contentlab.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png\",\"datePublished\":\"2020-03-24T19:28:26+00:00\",\"dateModified\":\"2023-03-03T19:45:32+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage\",\"url\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png\",\"contentUrl\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png\",\"width\":1746,\"height\":530},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/contentlab.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/contentlab.com\/#website\",\"url\":\"https:\/\/contentlab.com\/\",\"name\":\"Technical Content Marketing | ContentLab\",\"description\":\"Technical Content Marketing Done Right.\",\"publisher\":{\"@id\":\"https:\/\/contentlab.com\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/contentlab.com\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/contentlab.com\/#organization\",\"name\":\"ContentLab.io\",\"url\":\"https:\/\/contentlab.com\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/contentlab.com\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/favicon-1.png\",\"contentUrl\":\"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/favicon-1.png\",\"width\":768,\"height\":768,\"caption\":\"ContentLab.io\"},\"image\":{\"@id\":\"https:\/\/contentlab.com\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/x.com\/ContentLabIO\",\"https:\/\/www.linkedin.com\/company\/17893315\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\/\/contentlab.com\/#\/schema\/person\/e1b8cacc1019e83e8d6cd9ab1a049532\",\"name\":\"ContentLab\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/contentlab.com\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/3929e740ba8fe876f001a63220500e8e?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/3929e740ba8fe876f001a63220500e8e?s=96&d=mm&r=g\",\"caption\":\"ContentLab\"},\"description\":\"ContentLab provides high-quality written articles, tutorials, courses, and other technical marketing materials to industry leaders. We create no-nonsense tech content that\u2019s purpose-built to attract, educate, and engage your technical audience.\",\"sameAs\":[\"https:\/\/contentlab.op\"],\"url\":\"https:\/\/contentlab.com\/author\/contentlab\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/","og_locale":"en_US","og_type":"article","og_title":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab","og_description":"In this article we'll use Arm NN APIs to classify images of trash from a webcam attached to a Raspberry Pi.","og_url":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/","og_site_name":"Technical Content Marketing | ContentLab","article_published_time":"2020-03-24T19:28:26+00:00","article_modified_time":"2023-03-03T19:45:32+00:00","og_image":[{"width":1746,"height":530,"url":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","type":"image\/png"}],"author":"ContentLab","twitter_card":"summary_large_image","twitter_creator":"@ContentLabIO","twitter_site":"@ContentLabIO","twitter_misc":{"Written by":"ContentLab","Est. reading time":"12 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#article","isPartOf":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/"},"author":{"name":"ContentLab","@id":"https:\/\/contentlab.com\/#\/schema\/person\/e1b8cacc1019e83e8d6cd9ab1a049532"},"headline":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning","datePublished":"2020-03-24T19:28:26+00:00","dateModified":"2023-03-03T19:45:32+00:00","mainEntityOfPage":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/"},"wordCount":1864,"publisher":{"@id":"https:\/\/contentlab.com\/#organization"},"image":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage"},"thumbnailUrl":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","keywords":["Machine Learning","Neural Networks","Raspberry Pi"],"articleSection":["AI","Hands-on Tutorial","Portfolio"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/","url":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/","name":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning &#183; Technical Content Marketing | ContentLab","isPartOf":{"@id":"https:\/\/contentlab.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage"},"image":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage"},"thumbnailUrl":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","datePublished":"2020-03-24T19:28:26+00:00","dateModified":"2023-03-03T19:45:32+00:00","breadcrumb":{"@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#primaryimage","url":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","contentUrl":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","width":1746,"height":530},{"@type":"BreadcrumbList","@id":"https:\/\/contentlab.com\/automatic-trash-classification-with-raspberry-pi-and-arm-nn-machine-learning\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/contentlab.com\/"},{"@type":"ListItem","position":2,"name":"Automatic Trash Classification with Raspberry Pi and Arm NN Machine Learning"}]},{"@type":"WebSite","@id":"https:\/\/contentlab.com\/#website","url":"https:\/\/contentlab.com\/","name":"Technical Content Marketing | ContentLab","description":"Technical Content Marketing Done Right.","publisher":{"@id":"https:\/\/contentlab.com\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/contentlab.com\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/contentlab.com\/#organization","name":"ContentLab.io","url":"https:\/\/contentlab.com\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/contentlab.com\/#\/schema\/logo\/image\/","url":"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/favicon-1.png","contentUrl":"https:\/\/contentlab.com\/wp-content\/uploads\/2021\/11\/favicon-1.png","width":768,"height":768,"caption":"ContentLab.io"},"image":{"@id":"https:\/\/contentlab.com\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/ContentLabIO","https:\/\/www.linkedin.com\/company\/17893315\/"]},{"@type":"Person","@id":"https:\/\/contentlab.com\/#\/schema\/person\/e1b8cacc1019e83e8d6cd9ab1a049532","name":"ContentLab","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/contentlab.com\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/3929e740ba8fe876f001a63220500e8e?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/3929e740ba8fe876f001a63220500e8e?s=96&d=mm&r=g","caption":"ContentLab"},"description":"ContentLab provides high-quality written articles, tutorials, courses, and other technical marketing materials to industry leaders. We create no-nonsense tech content that\u2019s purpose-built to attract, educate, and engage your technical audience.","sameAs":["https:\/\/contentlab.op"],"url":"https:\/\/contentlab.com\/author\/contentlab\/"}]}},"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"https:\/\/contentlab.com\/wp-content\/uploads\/2019\/03\/4.png","_links":{"self":[{"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/posts\/3032"}],"collection":[{"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/comments?post=3032"}],"version-history":[{"count":39,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/posts\/3032\/revisions"}],"predecessor-version":[{"id":14077,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/posts\/3032\/revisions\/14077"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/media\/5331"}],"wp:attachment":[{"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/media?parent=3032"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/categories?post=3032"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/tags?post=3032"},{"taxonomy":"hashtags","embeddable":true,"href":"https:\/\/contentlab.com\/wp-json\/wp\/v2\/hashtags?post=3032"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}