PP-OCRv5 mobile models for NCNN

NCNN conversions of PaddlePaddle's PP-OCRv5 mobile text detector and script recognizers, used by translatte's on-device OCR. All credit for the models goes to PaddlePaddle / PaddleOCR; they are released under Apache-2.0, as are these conversions.

Directory	Source
`det`	PaddlePaddle/PP-OCRv5_mobile_det_onnx
`rec-cjk`	PaddlePaddle/PP-OCRv5_mobile_rec_onnx
`rec-latin`	PaddlePaddle/latin_PP-OCRv5_mobile_rec_onnx
`rec-eslav`	PaddlePaddle/eslav_PP-OCRv5_mobile_rec_onnx
`rec-korean`	PaddlePaddle/korean_PP-OCRv5_mobile_rec_onnx
`rec-arabic`	PaddlePaddle/arabic_PP-OCRv5_mobile_rec_onnx
`rec-devanagari`	PaddlePaddle/devanagari_PP-OCRv5_mobile_rec_onnx
`rec-th`	PaddlePaddle/th_PP-OCRv5_mobile_rec_onnx
`rec-ta`	PaddlePaddle/ta_PP-OCRv5_mobile_rec_onnx
`rec-te`	PaddlePaddle/te_PP-OCRv5_mobile_rec_onnx

Converted from the ONNX releases with pnnx 20260526 (convert_ocr_ncnn.py). Shapes are dynamic (detector: any HxW; recognizers: height 48, any width), weights are fp32, the input blob is in0 and the output out0. Preprocessing and the character dictionaries are the same as the ONNX models'; the dictionaries are not duplicated here.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support