bluecopa/paddleocr-v5-onnx

PaddleOCR v5 ONNX Models

PP-OCRv5 server models converted to ONNX for CPU inference via ort.

Models

File	Size	Purpose
PP-OCRv5_server_det_infer.onnx	88MB	Text region detection (DB++)
PP-OCRv5_server_rec_infer.onnx	81MB	Text recognition (server — printed + handwriting)
PP-LCNet_x1_0_textline_ori_infer.onnx	7MB	Text orientation classification
ppocrv5_dict.txt	74KB	Character dictionary (18,383 chars)

Input/Output

Detection: image [1, 3, H, W] → probability map [1, 1, H, W] Recognition: image [1, 3, 48, W] → logits [1, seq, 18385] (CTC decode with dict) Classification: image [1, 3, 48, 192] → rotation probability [1, 2]

Source

Converted from PaddlePaddle/PaddleOCR PP-OCRv5 server models.