Modern computer vision runs on neural networks that learn their own features from data instead of relying on hand-designed filters. Before moving on to specific vision tasks, it helps to understand how a network is built, how it learns, and how you train one in code.
A neural network is layers of simple units whose connections carry adjustable weights. Backpropagation works out how each weight contributed to the error, and gradient descent nudges the weights to reduce it. In PyTorch this becomes the training loop: a forward pass, a loss, a backward pass, and an optimizer step, repeated over your data.
For images, the key building block is the convolutional neural network, which slides learned filters across an image and pools the results into features a classifier can use. Watch the lessons in order, then continue on to Image Classification & Backbones.
A visual introduction to neural networks: layers, weights, and how a network can learn to recognize handwritten digits.
What happens inside a network as it learns: how weights are adjusted to reduce error, and an introduction to stochastic gradient descent.
One full training cycle in PyTorch: the forward pass, the loss, the backward pass, and the optimizer step.
How CNNs classify images step by step, using filters to build feature maps and pooling to condense them.