The image is passed through the image identifier to identify the schematic category of the image and the bounding box for the categorized object. A mask is used to identify the part of the image that represents the object. Given the foreground mask, the mask''s convex hull is placed and a minimally aligned aligned rectangle surrounding the hull is fitted. The aligned bounding box is rotated and resized, so that the foreground object is moved to approximately the standard orientation and size (referred to as being calibrated). The corrected image is used as the input of the granular categorization module to determine the granular category within the coarse category for the input image.
展开▼
