【无标题】 - 技术栈

works:

.dec/rec :parameters --> inference -->onnx
deal srt contrast
cls+ocr
flask
1+2 要求：图像输入 → 输出文本 + 文本框 + 文本比对
检测 ONNX：输入图像 → 输出文本框坐标
识别 ONNX（端到端内嵌字典 + 解码）：输入裁剪文本行图 → 直接输出字符串文本
上层业务：拿到框坐标 + 文本 → 执行文本比对

Start

1.1 从训练权重 → Paddle 推理模型 → ONNX 模型

bash 复制代码

python3 tools/export_model.py -c ./configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml -o Global.pretrained_model=./output/PP-OCRv5_mobile_det/first_train/best_accuracy.pdparams -o  Global.save_inference_dir=output/infer/PP-OCRv5_mobile_det_infer/first_det_0610
'''
(paddle) root@autodl-container-d8dlmpu3b9-2023ec6b:~/PaddleOCR-main# python3 tools/export_model.py -c ./configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml -o Global.pretrained_model=./output/PP-OCRv5_mobile_det/first_train/best_accuracy.pdparams -o  Global.save_inference_dir=output/infer/PP-OCRv5_mobile_det_infer/first_det
/root/miniconda3/envs/paddle/lib/python3.10/site-packages/paddle/utils/cpp_extension/extension_utils.py:715: UserWarning: No ccache found. Please be aware that recompiling all source files may be required. You can download and install ccache from: https://github.com/ccache/ccache/blob/master/doc/INSTALL.md
  warnings.warn(warning_message)
Skipping import of the encryption module.
W0610 15:23:09.306272  1993 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 8.9, Driver API Version: 13.0, Runtime API Version: 11.8
[2026/06/10 15:23:09] ppocr INFO: load pretrain successful from ./output/PP-OCRv5_mobile_det/first_train/best_accuracy
[2026/06/10 15:23:09] ppocr INFO: Export inference config file to output/infer/PP-OCRv5_mobile_det_infer/first_det/inference.yml
Skipping import of the encryption module
[2026/06/10 15:23:12] ppocr INFO: inference model is saved to output/infer/PP-OCRv5_mobile_det_infer/first_det/inference
'''

python 复制代码

python3 tools/export_model.py -c ./configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml -o Global.pretrained_model=./output/PP-OCRv5_mobile_rec/first/best_accuracy.pdparams -o  Global.save_inference_dir=output/infer/PP-OCRv5_mobile_rec_infer/first_rec_0610

(paddle) root@autodl-container-d8dlmpu3b9-2023ec6b:~/PaddleOCR-main# python3 tools/export_model.py -c ./configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml -o Global.pretrained_model=./output/PP-OCRv5_mobile_rec/first/best_accuracy.pdparams -o  Global.save_inference_dir=output/infer/PP-OCRv5_mobile_rec_infer/first_rec                                         Skipping import of the encryption module.
W0610 15:28:27.813534  2532 gpu_resources.cc:114] Please NOTE: device: 0, GPU Compute Capability: 8.9, Driver API Version: 13.0, Runtime API Version: 11.8
[2026/06/10 15:28:28] ppocr INFO: load pretrain successful from ./output/PP-OCRv5_mobile_rec/first/best_accuracy
[2026/06/10 15:28:28] ppocr INFO: Export inference config file to output/infer/PP-OCRv5_mobile_rec_infer/first_rec/inference.yml
Skipping import of the encryption module
[2026/06/10 15:28:31] ppocr INFO: inference model is saved to output/infer/PP-OCRv5_mobile_rec_infer/first_rec/inference

export_model.py生成ONNX

python 复制代码

import paddle
from paddle.static import InputSpec
from ppocr.modeling.architectures import build_model
from ppocr.postprocess import build_post_process
from ppocr.utils.save_load import load_pretrained_params
import yaml

# 1. 加载配置文件
config_path = './configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml'
with open(config_path, 'r') as f:
    config = yaml.safe_load(f)

# 2. 构建模型并加载权重
model = build_model(config['Architecture'])
load_pretrained_params(model, './output/PP-OCRv5_mobile_det/first_train/best_accuracy.pdparams')
model.eval()  # 切换到评估模式

# 3. 定义输入张量的形状（根据你的模型实际输入调整）
# OCR 检测模型通常输入为 [batch_size, channels, height, width]
input_spec = [InputSpec(shape=[1, 3, 640, 640], dtype='float32', name='image')]

# 4. 导出为 ONNX 格式
paddle.onnx.export(
    layer=model,
    path='./output/infer/PP-OCRv5_mobile_det_infer/first_det_0610/model',
    input_spec=input_spec,
    opset_version=13,  # 推荐使用 13 或更高版本以支持现代算子
    enable_onnx_checker=True
)

print("✅ ONNX 模型已成功导出！")

REC 导出ONNX

python 复制代码

import paddle
from paddle.static import InputSpec
from ppocr.modeling.architectures import build_model
from ppocr.utils.save_load import load_pretrained_params
# 导入官方的配置解析工具
from tools.program import get_config 

# 1. 使用官方工具加载配置（会自动读取你修改的字典路径并补全参数）
config_path = './configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml'
config, _, _ = get_config(config_path, show=False)

# 2. 构建模型并加载权重
model = build_model(config['Architecture'])
load_pretrained_params(model, './output/PP-OCRv5_mobile_rec/first/best_accuracy.pdparams')
model.eval()

# 3. 定义输入张量的形状（识别模型标准尺寸）
input_spec = [InputSpec(shape=[1, 3, 48, 320], dtype='float32', name='image')]

# 4. 导出为 ONNX 格式
paddle.onnx.export(
    layer=model,
    path='./output/infer/PP-OCRv5_mobile_rec_infer/first_rec_0610/model',
    input_spec=input_spec,
    opset_version=13,
    enable_onnx_checker=True
)

print("✅ 识别模型 ONNX 已成功导出！")

bash 复制代码