At Agira, Technology Simplified, Innovation Delivered, and Empowering Business is what we are passionate about. We always strive to build solutions that boost your productivity.

Image Classification Model – Create Your Own Using Javascript

  • By Nagaraj Ravi
  • September 29, 2020
  • 1344 Views

Image Classification Model helps you predict what is present in the image. If you are new to machine learning models or you don’t know in-depth about machine learning. Here you don’t need to worry about it because here we are going to create our object detection with our custom machine model in javascript using ML5.js and teachable machines. Creating the machine learning model with your data set of training data for your AI/ML project.

Step 1: (Create Model)

Go to a teachable machine site and create an image project for creating your image detection or image classification model.

Image Classification

Step 2: (Train data)

Now we need to create the categories and train the datasets. There are two categories(class) that will be default in teachable machines. I just simply created with those two categories if you want to more you can create it. After creating the categories you need to train the data by uploading the images respect to your class. Here I have trained the images for mobile and tv. After uploading the image click the train model button for training your model.

The accuracy depends on how much training data feeding to this model. I have just simply loaded a small amount of data.

Image Classification

Step 3: (Test the model)

After training the model completion. you need to test the model for how it gives the accuracy and how it detects the object from the image. So you need to test it by either webcam or upload it. Then, you can see the accuracy of your trained model.

Create your own Image Classification

Step 4: (Integrate it into your project)

 After the model is trained and tested successfully you need to export your model. for that you need to click the export model button. The export popup will open there and choose the tesorflow.js or p5.js option. Download and extract the zipped file in my_model folder. Finally, integrate your model in the image classifier function of ml5.j.

Image Classification

Also Read: What’s the Average JavaScript Developer Salary in the US?

Source code:

Index.html

<!DOCTYPE html>
<html>
   <head>
      <meta charset="UTF-8">
      <title>Image Classification Example</title>
      <script src="https://unpkg.com/ml5@0.3.1/dist/ml5.min.js" type="text/javascript"></script>
   </head>
   <body>
      <h1>Image classification using MobileNet</h1>
      <p>The MobileNet model labeled this as <span id="result">...</span> with a confidence of <span id="probability">...</span>.</p>
      <img src="" id="output_image" width="400" height="400" accept="image/*" crossorigin="anonymous" alt="Upload image">
      <form>
         <input type="file" id="file"  onchange="detectImage()">  
      </form>
      <script src="sketch.js"> 
      </script>
   </body>
</html>

Sketch.js

let classifier;
let labels = ['mobile','tv'];
preLoad();
// Initialize the Image Classifier method with your // Custom  model
function preLoad() {
classifier = ml5.imageClassifier('./my_model/model.json', modelLoaded);
}
function modelLoaded() {
console.log('Model Loaded!');
}
// predict the result after uploaded
function detectImage() {
var reader = new FileReader();
reader.onload = function () {
var output = document.getElementById('output_image');
output.src = reader.result;
classifier.classify(document.getElementById('output_image'), getResult);
}
reader.readAsDataURL(event.target.files[0]);
}
// result callback function
function getResult(err, results) {
alert(JSON.stringify(results));
        alert("Predicted is :", labels[results[0].label]);
}

Code Explanation

Initialize your label as per you created class in the teachable machine. It will be in the metadata.json.

let labels = ['mobile','tv'];

Load your model by setting the path in the ml5.imageClassifier method. It will be loaded while your app loads or when you call the preload function in your app.

classifier = ml5.imageClassifier('./my_model/model.json', modelLoaded);

After that when you upload the image it will show the result in the alert box.

Conclusion

Finally, we have come to the conclusion. By using the pre-trained model for detecting the image it won’t help you sometime when your requirement of detecting objects is not matching.  This blog will be helpful to you and enjoyed it. If I missed anything, you can comment below. Get the source code also available in GitHub.

Also Read: Top Open-Source JavaScript Frameworks for Front-End Web Development

Find the approximate cost to build your Application
Check Out

Nagaraj Ravi

A zealous full stack web developer with 3 years of professional experience. Hybrid application development and in developing Open-source projects are the best of my list. Out of passion ethical hacking is one of my hobbies. Yet, I am good at analyzing Web Security. And I teach machines with my expertise in Robotics and Machine learning.