PP-LCNet x1.0 Textline Orientation

PP-LCNet_x1_0_textline_ori is a text-line orientation classification model that distinguishes between 0-degree and 180-degree orientations of text lines, serving as a preprocessing stage in OCR pipelines to correct rotated inputs and improve downstream recognition accuracy. The model is based on the PP-LCNet_x0_25 backbone and classifies each text line into one of two classes: upright (0 degrees) or upside-down (180 degrees). It achieves a recognition average accuracy of 98.85% with a storage footprint of only 0.96 MB, making it suitable for lightweight deployment in document scanning, certificate photography, and other image-to-text workflows where captured images may be rotated. The table below summarizes the model's key metrics:

Model	Recognition Avg Accuracy (%)	Model Storage Size (M)	Introduction
PP-LCNet_x1_0_textline_ori	98.85	0.96	Text line classification model based on PP-LCNet_x0_25, with two classes: 0 degrees and 180 degrees

The following image shows an example output after classification, with the text line identified as rotated 180 degrees: The model is licensed under Apache 2.0. For further background on the OCR toolkit from which it originates, see the PaddleOCR repository.