Recognizing Figure Labels in Patents [AAAI 2021 SDU Workshop]

* 100patents_design_original_png.zip: 100 figures extracted from 100 US DESIGN patents, original (unrotated), PNG format * 100patents_design_original-tif.zip: 100 figures extracted from 100 US DESIGN patents, original (unrotated), TIF format* 100patents_design_rotated-tif.zip: 100 figures extracted from 100 US DESIGN patents, rotated to upright if needed TIF format* 100patents_design_rotated-png.zip: 100 figures extracted from 100 US DESIGN patents, rotated to upright if needed, PNG format* all_results_labels.xml: ground truth labels and labels extracted by 8 tools: SWT, Adobe Acrobat, EAST, Amazon Textract, Tesseract, Google Vision API, Abbyy, and the alpha-shape method.
For details, see the following paper: Gong Ming, Xin Wei, Diane Oyen, Jian Wu, Martin Gryder, and Liping Yang. "Recognizing Figure Labels in Patents." In: AAAI-2021 workshop on Scientific Document Understanding (SDU). Virtual Event.