量化的飞桨模型部署正常，但是转化为onnx模型出现问题（非量化的模型转为onnx可以正常部署）

[stringency]

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

• 系统环境/System Environment：python3.10.4
• 版本号/Version：Paddle： PaddleOCR： 问题相关组件/Related components：
  paddle2onnx 1.0.6
  paddlepaddle 2.5.2
  paddleslim 2.6.0
• 运行指令/Command Code：

• 微调量化onnx部署

python tools/infer/predict_system.py --use_onnx=True --use_gpu=False --det_model_dir=output/CCPD/det_quant/onnx/model.onnx --rec_model_dir=output/CCPD/rec_quant/onnx/model.onnx --image_dir="F:/DataSet/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg" --rec_image_shape=3,48,320

• 完整报错/Complete Error Message：
  [2024/05/18 02:21:46] ppocr INFO: In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320
  [2024/05/18 02:21:46] ppocr DEBUG: dt_boxes num : 0, elapsed : 0.13833069801330566
  [2024/05/18 02:21:46] ppocr DEBUG: rec_res num : 0, elapsed : 0.0
  [2024/05/18 02:21:46] ppocr DEBUG: 0 Predict time of F:/DataSet/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg: 0.141s
  [2024/05/18 02:21:46] ppocr DEBUG: The visualized image saved in ./inference_results\04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg
  [2024/05/18 02:21:46] ppocr INFO: The predict total time is 0.1968538761138916

[stringency]

为了对比我提供飞桨模型的运行结果

飞桨模型非onnx模型的微调量化部署

python tools/infer/predict_system.py --det_model_dir=output/CCPD/det_quant/infer/ --rec_model_dir=output/CCPD/rec_quant/infer/ --image_dir="F:/DataSet/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg" --rec_image_shape=3,48,320

运行结果

[2024/05/18 02:34:52] ppocr INFO: In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320
[2024/05/18 02:34:52] ppocr DEBUG: dt_boxes num : 1, elapsed : 0.12008166313171387
[2024/05/18 02:34:52] ppocr DEBUG: rec_res num : 1, elapsed : 0.008484840393066406
[2024/05/18 02:34:52] ppocr DEBUG: 0 Predict time of F:/DataSet/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg: 0.133s
rec_image_shape=3,48,320
E0518 02:37:38.138892 5744 analysis_config.cc:121] Please use PaddlePaddle with GPU version.
E0518 02:37:38.592444 5744 analysis_config.cc:121] Please use PaddlePaddle with GPU version.
[2024/05/18 02:37:38] ppocr INFO: In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320
[2024/05/18 02:37:39] ppocr DEBUG: dt_boxes num : 1, elapsed : 1.078622579574585
[2024/05/18 02:37:40] ppocr DEBUG: rec_res num : 1, elapsed : 0.1730790138244629
[2024/05/18 02:37:40] ppocr DEBUG: 0 Predict time of F:/DataSet/CCPD2020/ccpd_green/test/04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg: 1.254s
[2024/05/18 02:37:40] ppocr DEBUG: 皖AD867660, 0.873
[2024/05/18 02:37:40] ppocr DEBUG: The visualized image saved in ./inference_results\04131106321839081-92_258-159&509_530&611-527&611_172&599_159&509_530&525-0_0_3_32_30_31_30_30-109-106.jpg
[2024/05/18 02:37:40] ppocr INFO: The predict total time is 1.304152488708496

[Zheng-Bicheng]

幸苦提供一下量化以后的PaddleOCR模型，我这边测试一下

[stringency]

幸苦提供一下量化以后的PaddleOCR模型，我这边测试一下

output/CCPD/det_quant/onnx/model.onnx:
det_quant_model.zip

output/CCPD/rec_quant/onnx/model.onnx:
rec_quant_model.zip

[Homura852]

你好，请问你在将检测模型进行量化训练时，训练速度如何呢？为什么我这边正常训练模型速度很快，但是转为量化训练就会慢很多。

[github-actions]

This issue is stale because it has been open for 30 days with no activity.

[stringency]

您好，您发的邮件已经收到，谢谢！

[github-actions]

This issue is stale because it has been open for 30 days with no activity.

[stringency]

您好，您发的邮件已经收到，谢谢！