FedPrivacyMask Dataset: Privacy-Annotated Logistics Waybills for Privacy Region Segmentation

This dataset is a self-constructed, privacy-annotated simulated logistics waybill dataset for privacy region segmentation research. Dataset Contents: 20,997 PNG images (processed_*.png), including original self-designed waybills and augmented versions (random occlusion, rotation) 20,997 CSV annotation files (stored in the CSV/ subfolder), one-to-one correspondence with the images CSV File Columns: reading_order: reading order of characters on the waybill character: the recognized character X_coordinate, Y_coordinate: position of the character in the image width, height: size of the character confidence: OCR recognition confidence score is_privacy: whether the character belongs to a privacy region privacy_type: phone / name / address remarks: additional notes for phone number characters Data Construction Notes: All waybill templates were designed by the authors Names, phone numbers, and addresses follow real-world formatting rules (e.g., 11-digit phone numbers, province/city/district/street address hierarchy) but are all synthetically generated Sensitive digits in phone numbers have been anonymized (replaced with # or *) The dataset contains no real personal information and does not correspond to any real logistics order This dataset can be publicly shared without privacy risks License: CC BY 4.0 Contact: [e-mail]