The Difference Between Segmentation and Per Pixel Classification

We're taking this continuous data imagery of the world probably and we're trying to find discrete objects in it. So when we train the machine, we're labelling objects that we can see in the imagery. Are we drawing bounding boxes around these objects or do we have to draw very tight, detailed polygons around them? Is that what the process looks like? Yeah so what you do on the platform is you draw a somewhat detailed polygon. And the model it will not only look at the pixel so it's not a per pixel classification, it will also look at the context and the model will output the polygon.