The CV Task Landscape

So far you've met CNNs as classifiers: one image in, one label out. But real vision systems ask richer questions, and each gets its own task:

• Classification — what is in this image? One label.
• Object detection — what is here, and where? A bounding box plus a label per object.
• Segmentation — which pixels belong to what? A mask covering every object.
• Tracking — is this the same object as last frame? Identity carried over time.

The remarkable part: all four reuse the same backbone — the convolutional feature extractor you already know. What changes is the head bolted on top and the loss it's trained with.