Teach your computer to recognize images, sounds and movements (without coding!)

Training ML models is not easy for the average user. Machine learning and artificial intelligence is a relatively new area that most computer users have not addressed. Of course, not everyone is interested in programming, but even that should not prevent individuals from embarking on these waters. That's why Google launched the "Teachable Machine" project.

What is ML?

Literally translated, it means "Training the machine", ie training the computer to recognize something. For example, you can "feed" a model with hundreds of images of an apple, let the algorithm do its thing, and later it will be able to recognize any apple. Read more about ML here.

What is a Teachable Machine?

It is a web application that has a very nice and modern design, and it was created so that anyone who knows how to use the keyboard and mouse can create their own ML model.

This project was created so that anyone, even those who had no contact with programming or were not even interested in computers, could create their own ML model and later use it in some of their projects.

So, as I wrote in the ML paragraph, you have to "feed" the model with pictures of something to "learn" what that thing looks like in multiple variations. This application has simplified the whole process and brought ML closer to end-users.

What is even easier is that the user does not have to learn the model from scratch what is a contour, what is a color, an animal, or a dog… but can only teach what a particular dog looks like, or something fifth. This is possible thanks to Google's great model, which has already "learned" what everything in the world looks like, so you can "take" what you need from it (what your dog looks like, for example).

So you can, for example, take a few pictures of yourself, your dog, and your cat via a webcam, let the app "learn", and then test whether it recognizes what's on the camera at some point.

In addition to training the model with images, you can also train it with movements and sound. For example, you can train him to recognize when you do a squat and count or what kind of music you play.

Instruction

1. Open the project site - link
2. Choose whether you want to train the model with pictures, sound, or movement
3. Add training files to each class (you can add via webcam or choose from your computer)
4. Click "Train model"
5. Test!

Once you test your model and are satisfied with the results, you can report it and continue to use it in your projects. You can find good examples on the official website of the project.

Advanced

When you open the tool you will notice that there are some classes in the first step. This is useful if you want to train a model with multiple objects, sounds, or movements. For example, you can send him shots of apples, pears, and cherries. Place each fruit in one class. Later when you test you will see how much the model is sure of which fruit is on the screen at some point.

If you open the advanced options in the second step (Training) you will see settings for epochs, amount, and rate of learning. These are fine-tuned learning models and I recommend you leave the default. You can adjust this only when you understand how ML really works and you want to fine-tune the results. If you have a lot more classes, these options can help you make a more precise model.

You can use your trained model in some of your projects, and TensorFlow, ML5js, p5.js, Coral.ai, Framer, and Node.js are officially supported.