@@ -16,8 +16,64 @@ comments: true
1616
1717该步骤主要先基于PaddleX可以正确使用PP-OCRv4_server_rec_doc模型得到正确结果。
1818
19+ 该部分主要参考文档:[ docs] ( https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html#_3 )
20+
21+ 安装` paddlex ` :
22+
23+ ``` bash linenums="1"
24+ pip install " paddlex[ocr]==3.0.0rc1"
25+ ```
26+
27+ 测试PP-OCRv4_server_rec_doc模型能否正常识别:
28+
29+ 测试用图:
30+
31+ ![ alt text] ( ../images/1.jpg )
32+
33+ !!! tip
34+
35+ 运行以下代码时,模型会自动下载到**/Users/用户名/.paddlex/official_models**下。
36+
37+ ``` python linenums="1"
38+ from paddlex import create_model
39+
40+ model = create_model(model_name = " PP-OCRv4_server_rec_doc" )
41+ output = model.predict(input = " images/1.jpg" , batch_size = 1 )
42+ for res in output:
43+ res.print()
44+ res.save_to_img(save_path = " ./output/" )
45+ res.save_to_json(save_path = " ./output/res.json" )
46+
47+ # 输出以下内容,表明成功:
48+ # {'res': {'input_path': 'images/1.jpg', 'page_index': None, 'rec_text': '绿洲仕格维花园公寓', 'rec_score': 0.9839767813682556}}
49+ ```
50+
1951#### 1. 模型转换
2052
53+ 该部分主要参考文档: [ docs] ( https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html?h=paddle2onnx#22 )
54+
55+ PaddleX官方集成了paddle2onnx的转换代码:
56+
57+ ``` bash linenums="1"
58+ paddlex --paddle2onnx --paddle_model_dir models/PP-OCRv4_server_rec_doc --onnx_model_dir models/PP-OCRv4_server_rec_doc
59+ ```
60+
61+ 输出日志如下,表明成功:
62+
63+ ``` bash linenums="1"
64+ Input dir: models/PP-OCRv4_server_rec_doc
65+ Output dir: models/PP-OCRv4_server_rec_doc
66+ Paddle2ONNX conversion starting...
67+ warnings.warn(warning_message)
68+ [Paddle2ONNX] Start parsing the Paddle model file...
69+ [Paddle2ONNX] Use opset_version = 7 for ONNX export.
70+ [Paddle2ONNX] PaddlePaddle model is exported as ONNX format now.
71+ 2025-05-14 08:21:23 [INFO] Try to perform optimization on the ONNX model with onnxoptimizer.
72+ 2025-05-14 08:21:23 [INFO] ONNX model saved in models/PP-OCRv4_server_rec_doc/inference.onnx.
73+ Paddle2ONNX conversion succeeded
74+ Done
75+ ```
76+
2177#### 2. 模型推理验证
2278
2379#### 3. 模型精度测试
0 commit comments