# PP-OCRv4 text detector

PaddleOCR PP-OCRv4 mobile DB text detection model converted to ONNX and distributed by SWHL/RapidOCR.

Source: https://huggingface.co/SWHL/RapidOCR/tree/1cfba2e90fc938db55889873735088de210cc173/PP-OCRv4

File: ch_PP-OCRv4_det_infer.onnx

License: Apache-2.0. See PaddleOCR-Apache-2.0.txt.

Size: 4,745,517 bytes

SHA256: d2a7720d45a54257208b1e13e36a8479894cb74155a5efe29462512d42f49da9

ClearFrame uses BGR NCHW preprocessing, max-side 960px, per-channel mean [0.485,0.456,0.406], std [0.229,0.224,0.225]. Model weights are unmodified. ClearFrame's connected-component postprocessing, horizontal-line grouping and bounded rectangle expansion are its own implementation, rather than the upstream OpenCV/Vatti implementation. This is text-region detection without a recognition model or semantic subtitle classifier.