Given its produced by a ".ai" company I would imagine they are using a U-Net derived ML model and not traditional OpenCV methods. You can read more about that

Given its produced by a ".ai" company I would imagine they are using a U-Net derived ML model and not traditional OpenCV methods. You can read more about that approach here: https://hackernoon.com/releasing-supervisely-person-dataset-for-teaching-machines-to-segment-humans-1f1fc1f28469 https://hackernoon.com/releasing-supervisely-person-dataset-... Good results though for sure and very handy for quick image edits.