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.
