Convert Datumaro person annotations to Head BBox JSON format with occlusion data
• Structure: items (each with image and annotations), nested categories.label.labels
• Geometry: Supports points (polygons) and bbox (bounding boxes)
• Categories: Nested structure with categories.label.labels array containing label definitions
• Label References: Annotations use label_id to reference category indices
• image_id: Unique identifier for the image
• filename: Image file path
• width, height: Image dimensions
• total_count: Number of person annotations found
• annotations: Array of person bounding boxes with id, bbox [x, y, width, height], and occlusion (0 or 1)
• Bounding box format: Top-left coordinates (x, y) with width and height in pixels
• Only person annotations are included
• Datumaro → Head BBox: Converts person annotations to bounding box format
• Polygon Conversion: Converts polygon points to bounding box coordinates
• Person Filtering: Only includes annotations with person labels
• Single Image: Processes the first item from Datumaro data
• Coordinate Preservation: Maintains top-left based coordinates from Datumaro
• Occlusion Transfer: Maps occluded attribute from Datumaro (false=0, true=1)