PP FormulaNet plus L Introduction PP FormulaNet plus is an enhanced version of the formula recognition model developed by the PaddleOCR Team, building upon the original PP FormulaNet. Compared to the original version, PP FormulaNet plus utilizes a more diverse formula dataset during training, including sources such as Chinese dissertations, professional books, textbooks, exam papers, and mathematics journals. This expansion significantly improves the model’s recognition capabilities. The PP FormulaNet plus includes multiple versions: L, M, and S, where PP FormulaNet plus L have added support for Chinese formulas and increased the maximum number of predicted tokens for formulas from 1,024 to 2,560, greatly enhancing the recognition performance for complex formulas. The key accuracy metrics are as follow: Model Backbone En BLEU↑ Zh BLEU(%)↑ GPU Inference Time (ms) : : : : : : : : : : UniMERNet Donut Swin 85.91 43.50 2266.96 PP FormulaNet S PPHGNetV2 B4 87.00 45.71 202.25 PP FormulaNet L Vary VIT B 90.36 45.78 1976.52 PP FormulaNet plus S PPHGNetV2 B4 88.71 53.32 191.69 PP FormulaNet plus M PPHGNetV2 B6 91.45 89.76 1301.56 PP FormulaNet plus L Vary VIT B 92.22 90.64 1745.25 LaTeX OCR…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy