An image classifier looks at a picture and decides what it shows: a cat or a dog, a handwritten digit, a type of product. Instead of hand-written rules, a convolutional neural network (CNN) learns visual patterns such as edges, textures and shapes directly from labeled example images.
The lessons start with a plain-language explanation of how CNNs work. Next you build a cat vs dog classifier in Keras, then train a CNN on the small-image CIFAR-10 dataset. The last lesson covers transfer learning with a pretrained ResNet-50, so you can get strong results on your own images without training from scratch.
Watch the lessons in order and code along in Google Colab or a Jupyter notebook. A GPU runtime in Colab makes training noticeably faster. If a video runs long, treat it as a reference you can dip back into while you build your own version.
A beginner-friendly explanation of CNNs with very little math: how filters detect features, what convolution does, and why ReLU and pooling matter.
VideoBuild a cat vs dog classifier with a CNN in Keras: load the dataset, define and train the model, then predict the class of individual images.
Video 18minClassify small images with TensorFlow and Keras: download and visualize the CIFAR-10 dataset, then train a convolutional neural network on it.
VideoReuse the weights of a pretrained CNN such as ResNet, Inception or VGG and fine-tune it for your own image classification problem.
Video