We have a technology can automatically generate training image data sets for deep learning with true contour annotations, i.e. objects labeled without background. Instead of conventional bounding boxes around annotated objects that contain irrelevant background in the corners, we can provide training images that are cropped to the contour of an object of interest with >90% relevant fill factor (either non-rectangular ROI or rectangular ROI with zero background outside the object contour).
What are the use cases for such data? What benefits can it offer?