Google’s Teachable Machine service makes it possible for anyone to experiment with machine learning at home and create small, useful applications without much programming knowledge or cost. In this article, we explore how to create a model that recognizes objects in images. Specifically, we focus on distinguishing between dogs, cats, lions, and birds, with the AI also classifying bird species not included in the training data set.
To successfully train a model, it’s important to collect enough footage. At least 200 photos are usually required for each object to be recognized, and the more templates available, the better the results. The training can be performed on a normal PC; mini computers like the Raspberry Pi are too weak for this task but can still be used once the model has been pre-trained.
We demonstrate how to export the finished model in Tensorflow Lite format and run it on the Raspberry Pi, where any photos of animals can be taken using a Raspi camera and the AI will attempt to classify them. The same process can be applied to other areas, such as audio or gesture recognition, and with enough training images, even face recognition can be implemented.
With this approach, users can experiment with machine learning and create small, practical applications while building their knowledge and skills in the field. This is now more accessible than ever, making it possible for individuals and smaller organizations to develop AI models without the resources and investment of larger corporations like OpenAI.
