Image annotation is the process of labeling images, typically to create a ground truth dataset for training or evaluating computer vision models. Annotation can take many forms: assigning a single label to an entire image (image classification), drawing bounding boxes around objects (object detection), creating pixel-level masks for regions (segmentation), marking keypoints (pose estimation, facial landmarks), or adding descriptive captions (image captioning). This is usually done manually by humans using specialized tools, although there’s also AI-assisted labeling to speed it up. Quality annotations are crucial – they need to be accurate and consistent. Large annotated datasets (like COCO, Pascal VOC, ImageNet) have fueled advancements in CV. The difficulty and cost of image annotation often dictate what tasks are feasible; for example, pixel-wise segmentation is much more labor-intensive than tagging an image with a label.