Extract Features is using detectors that are trained with machine learning, to automatically extract information from your photos, point clouds and meshes.
You can start extracting features from data that is part of your current iTwin Capture Modeler project (e.g. photos, block, point cloud productions, reality meshes).
Alternatively, you can also Extract Features on reality data that are not within your iTwin Capture Modeler project. These are more advanced workflows and require learning about ContextScene format.