PaddleOCR安装报错

[DazzlingGalaxy]

系统CentOS 7.9.2009 x64，conda新建的虚拟环境，python 3.8.16，pip源是阿里源。

我用下面的命令安装

pip install paddlepaddle-gpu==2.4.2.post112 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
pip install paddleocr

paddlepaddle安装成功，paddleocr安装报错

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
paddlepaddle-gpu 2.4.2.post112 requires protobuf<=3.20.0,>=3.1.0, but you have protobuf 3.20.3 which is incompatible.

此时查看版本是
paddleocr==2.6.1.3
paddlepaddle-gpu==2.4.2.post112

然后我卸载protobuf 3.20.3再安装3.20.0，报错

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
onnx 1.13.1 requires protobuf<4,>=3.20.2, but you have protobuf 3.20.0 which is incompatible.

再卸载onnx和protobuf，再安装protobuf==3.20.0(发布于2022年4月2日)和onnx==1.12.0(发布于2022年6月18日，上个版本1.11.0发布于2022年2月18日)。由于装onnx时会自动装上protobuf导致protobuf装的版本高，所以先手动装protobuf再装onnx。都装好后测试了一次可以OCR识别(第一次运行时自动下载了几个文件)，但测试第二次时报错

ERROR in app: Exception on /bankticketocr [POST]
Traceback (most recent call last):
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/flask/app.py", line 2528, in wsgi_app
    response = self.full_dispatch_request()
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/flask/app.py", line 1825, in full_dispatch_request
    rv = self.handle_user_exception(e)
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/flask/app.py", line 1823, in full_dispatch_request
    rv = self.dispatch_request()
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/flask/app.py", line 1799, in dispatch_request
    return self.ensure_sync(self.view_functions[rule.endpoint])(**view_args)
  File "sfz_ocr_test.py", line 799, in bankticket
    return bank_ticket_ocr(img_file_path=img_file_path, shoukuan_name=shoukuan_name, shoukuan_bank_number=shoukuan_bank_number, shoukuan_bank_name=shoukuan_bank_name)
  File "sfz_ocr_test.py", line 709, in bank_ticket_ocr
    ocr = PaddleOCR(enable_mkldnn=True, use_tensorrt=True, use_angle_cls=False, lang="ch")
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/paddleocr/paddleocr.py", line 523, in __init__
    super().__init__(params)
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/paddleocr/tools/infer/predict_system.py", line 46, in __init__
    self.text_detector = predict_det.TextDetector(args)
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/paddleocr/tools/infer/predict_det.py", line 141, in __init__
    self.predictor, self.input_tensor, self.output_tensors, self.config = utility.create_predictor(
  File "/root/miniconda3/envs/ocr_gpu/lib/python3.8/site-packages/paddleocr/tools/infer/utility.py", line 277, in create_predictor
    predictor = inference.create_predictor(config)
RuntimeError: (NotFound) TensorRT is needed, but TensorRT dynamic library is not found.
  [Hint: dso_handle should not be null.] (at /paddle/paddle/fluid/platform/dynload/tensorrt.cc:47)

总结一下，两个问题：
1.为什么使用下面的命令安装时报错？

pip install paddlepaddle-gpu==2.4.2.post112 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
pip install paddleocr

2.为什么我手动更换protobuf和onnx的版本后，第一次能识别图片中的文字，第二次就报错？

[andyjiang1116]

检查下tensorrt环境配置，报错提示是tensorRT问题，或者先关闭tensorrt试试

[DazzlingGalaxy]

检查下tensorrt环境配置，报错提示是tensorRT问题，或者先关闭tensorrt试试

1.问题一报错的原因是什么？你们新环境安装会报这个错吗？
2.新买的物理服务器，在这之前就装过GPU版的PaddleSpeech，还是在另一个conda环境中用pip装的，怎么检查tensorrt环境配置？我是新手。
3.怎么关闭tensorrt？

[andyjiang1116]

安装问题可能是python版本，paddle版本，cuda版本的不匹配问题，通过use_tensorrt=False关闭

[DazzlingGalaxy]

安装问题可能是python版本，paddle版本，cuda版本的不匹配问题，通过use_tensorrt=False关闭

1.应该不是版本不匹配的问题吧？因为我的PaddleSpeech正常，而且PaddleOCR的安装说明里也用的python 3.7和3.8。所以安装报错到底什么原因？你们可以在新环境试一下吗？

2.use_tensorrt无论是True还是不写，我发现识别速度差别不大啊，那关闭它有什么影响？

3.先说一下我的服务器，CPU是Intel Xeon Gold 6138，GPU是Tesla P4，之前在服务器报错的代码是这样的

def ocr_sfz(img_path):
    ocr = PaddleOCR(enable_mkldnn=True, use_tensorrt=True, use_angle_cls=False, lang="ch")
    result = ocr.ocr(img_path, cls=True)

def bank_ticket_ocr(img_file_path: str, shoukuan_name: str, shoukuan_bank_number: str, shoukuan_bank_name: str):
    ocr = PaddleOCR(enable_mkldnn=True, use_tensorrt=True, use_angle_cls=False, lang="ch")
    result = ocr.ocr(img_path, cls=True)

最开始写这些代码时还没有带独显的服务器，用的是CPU版的PaddleOCR，测试管用后才放在现在带P4的服务器上，并且换成了GPU版的PaddleOCR。上面两个函数对应两个接口，最开始我说第一次成功识别，第二次报错，是否因为第二次时重复初始化了PaddleOCR()对象？因为我想起来用PaddleSpeech是看到这两个链接：PaddlePaddle/PaddleSpeech#2881
PaddlePaddle/PaddleSpeech#2908

4.今天我又改了代码测试了下，不报错了。相对之前报错代码的变化，一是将PaddleOCR()对象初始化为全局变量，二是删除enable_mkldnn=True

ocr = PaddleOCR(use_tensorrt=True, use_angle_cls=False, lang="ch")

def ocr_sfz(img_path):
    result = ocr.ocr(img_path, cls=True)

def bank_ticket_ocr(img_file_path: str, shoukuan_name: str, shoukuan_bank_number: str, shoukuan_bank_name: str):
    result = ocr.ocr(img_path, cls=True)

4.1 是不是将PaddleOCR()对象初始化为全局变量，比在每个函数中初始化一次要好？
4.2 我现在服务器有P4，装的也是GPU版的PaddleOCR，还有必要开启enable_mkldnn吗？

[andyjiang1116]

将PaddleOCR()对象初始化为全局变量，比在每个函数中初始化一次要好

是的，全局变量要好一些，enable_mkldnn这个是在cpu下开启才有用

[DazzlingGalaxy]

更新(以下内容仅在Windows测试过)：
不用先卸载onnx和protobuf再重新安装。
paddleocr安装报错后直接执行：

pip install protobuf==3.20.0
pip install onnx==1.12.0

就能覆盖掉版本，装上3.20.0和1.12.0。

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.

[Kael-DWT]

解决过程：
报错： ImportError: cannot import name '_print_arguments' from 'paddle.distributed.utils' (/home/kael.yang/.local/lib/python3.8/site-packages/paddle/distributed/utils/init.py)

pip3 install --upgrade pip
python3 -m pip install paddlepaddle-gpu==2.1.1 -i https://mirror.baidu.com/pypi/simple
pip3 install -U https://paddleocr.bj.bcebos.com/whl/layoutparser-0.0.0-py3-none-any.whl
pip install "paddleocr>=2.2"

这里在ipython测试: from paddleocr import PPStructure
报错：
TypeError: Descriptors cannot not be created directly.
If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.
If you cannot immediately regenerate your protos, some other possible workarounds are:

1. Downgrade the protobuf package to 3.20.x or lower.
2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

然后解决：
pip install protobuf==3.20.0

最后成功可使用 from paddleocr import PPStructure

---

还是没有成功， 最后解析时还是报错：

C++ Traceback (most recent call last):

0 paddle::framework::SignalHandle(char const*, int)
1 paddle::platform::GetCurrentTraceBackStringabi:cxx11

---

Error Message Summary:

FatalError: Segmentation fault is detected by the operating system.
[TimeInfo: *** Aborted at 1691053082 (unix time) try "date -d @1691053082" if you are using GNU date ***]
[SignalInfo: *** SIGSEGV (@0x0) received by PID 3901491 (TID 0x7f80863a7740) from PID 0 ***]

段错误