how to make exe with pyinstaller

[Heryk13]

how to make exe from python file with paddleOCR

i am using auto-py-to-exe but i think it will work in pyinstaller too with the same parameters to collect data.

install auto-py-to-exe

for installing auto-py-to-exe
$ pip install auto-py-to-exe
Then to run it
$ auto-py-to-exe

github page: https://github.com/brentvollebregt/auto-py-to-exe

select you main file .py and add the paddle dependencies in the Advanced options

it take me a while to discover this.
aq é brasil HUE HUE HUE HUE BR

#6875 #9951

[pikut]

In my case, script return "ModuleNotFoundError: No module named 'scipy.io'", but other problems gone - your solution fix it.

[Heryk13]

in that case if you add in the --hidden-import scipy i think it will fix too

[pikut]

Unfortunely does not work, the script return the same error. What kind of python version you use?

[Heryk13]

I am using 3.10.13, i tried with the 3.11 but i dont remember why it didn't worked

If in hidden import is not working, try inserting the scipy in collect-data or colect-all

[alejandroNoc]

For me it work with this , im trying to do with PPOCR after following and exploring finally i got it:
hope in somewhile this help to other persons

[Edmond-Lee-Zse-Wong]

I still find an error after using your method, maybe I should add more file?

[Edmond-Lee-Zse-Wong]

After adding scipy and lmdb, unfortunately it still doesn't work. Below are my codes, written in the simplest way possible.

import traceback
import time

try:
    from paddleocr import PaddleOCR
    ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # need to run only once to download and load model into memory
    img_path = 'img5.png'
    result = ocr.ocr(img_path, cls=True)
    for idx in range(len(result)):
        res = result[idx]
        for line in res:
            print(line)
except Exception as e:
    traceback.print_exc()
    time.sleep(60)

After that is the error information

Traceback (most recent call last):
  File "test.py", line 7, in <module>
  File "PyInstaller\loader\pyimod02_importers.py", line 419, in exec_module
  File "paddleocr\__init__.py", line 14, in <module>
    from .paddleocr import *
  File "PyInstaller\loader\pyimod02_importers.py", line 419, in exec_module
  File "paddleocr\paddleocr.py", line 48, in <module>
    from tools.infer import predict_system
  File "E:\Work_Projects\Intelligence_Group\Overall_Venv\venv\Scripts\output\test\_internal\paddleocr\tools\infer\predict_system.py", line 33, in <module>
    import tools.infer.predict_det as predict_det
  File "E:\Work_Projects\Intelligence_Group\Overall_Venv\venv\Scripts\output\test\_internal\paddleocr\tools\infer\predict_det.py", line 31, in <module>
    from ppocr.data import create_operators, transform
  File "E:\Work_Projects\Intelligence_Group\Overall_Venv\venv\Scripts\output\test\_internal\paddleocr\ppocr\data\__init__.py", line 37, in <module>
    from ppocr.data.lmdb_dataset import LMDBDataSet, LMDBDataSetSR, LMDBDataSetTableMaster
  File "E:\Work_Projects\Intelligence_Group\Overall_Venv\venv\Scripts\output\test\_internal\paddleocr\ppocr\data\lmdb_dataset.py", line 17, in <module>
    import lmdb
ModuleNotFoundError: No module named 'lmdb'

I‘ve already added scipy and lmdb like you @pikut say

I have no idea where the problem lies. /(ㄒoㄒ)/~~

• System Environment：win10
• Version：Paddle：2.5.2 PaddleOCR：2.7.0.3 Related components：python：3.10.1 pyinstaller：6.3.0

[Heryk13]

@Edmond-Lee-Zse-Wong is there any specific reason to use the python version 3.10.1? i am using anaconda, if you are going to use the python 3.10 i recommend using a more newer version of 3.10, maybe its the version of python you are using. i will test tomorrow in a newer version of 3.10 and in the 3.10.1. recommend using anaconda too

[Edmond-Lee-Zse-Wong]

No, there is no specific reason to use 3.10.1. I'm just unsure which version I should choose. Previously I used 3.7.6, but Paddle doesn't support this version, so I switched to whichever version would make it runs. Thanks very much for your @Heryk13 passion and patience!

[Heryk13]

hey @Edmond-Lee-Zse-Wong i created a exe with your code. here is what i used

code:
`import traceback
import time

try:
from paddleocr import PaddleOCR
ocr = PaddleOCR(use_angle_cls=True, lang="ch") # need to run only once to download and load model into memory
img_path = "img.png"
result = ocr.ocr(img_path, cls=True)
for idx in range(len(result)):
res = result[idx]
for line in res:
print(f"ocrText: {line[1][0]}")
except Exception as e:
traceback.print_exc()
time.sleep(60)
`

i tested in the python versions 3.10.13 and i saw the error scipy, and did what @pikut said.
like this.

and it worked i think you could try in the python version 3.10.13 or 3.9.13 and 3.10.11 that worked for @pikut too.

the requirements
requirements.txt

extra:
i tested with 3.11.7 too but it didn't worked because PyMuPDF, i tried this solutions but no luck too.
https://stackoverflow.com/questions/76379293/how-can-i-fix-the-error-in-pymupdf-when-installing-paddleocr-with-pip

[Edmond-Lee-Zse-Wong]

Finally it works!!! Thank you@Heryk13 @pikut so much!!!

[Heryk13]

im glad to hear that 👍

[pikut]

That's cool!

[SubhadeepBanerjeeChowdhury]

Thank you so much, this worked!

[NguyenDucQuan12]

hi, i am using yolov8 along with paddleocr to read detected characters. I want to wrap it all up, I see you are also using paddleocr, ultralytics so wanted to ask for your opinion. When I repackage it, when I install it to another device (without any modules), it cannot run. I want all the modules to be repackaged and the user can just install the software. use. Can you give me your opinion? Thank you

[Heryk13]

i think you need to multi thread the video update itself, because i cant drag the window because the update of the video.

[NguyenDucQuan12]

I will update the UI using Qt, I also tried multithreading for video streaming from the camera, however it seems to be having some errors, it doesn't actually show or notify the problem so I can't solve it. I think it's in cv2

[NguyenDucQuan12]

@Heryk13 Do you have any suggestions for using multithreading in this case

[Heryk13]

if you are going to use the qt there is the threading of the qt you can use with is very good and flexible to a lot of things
Qthreads its the name of the thread in qt and you can use the qt designer wich can be very easy to draw UI.
https://build-system.fman.io/qt-designer-download

you will write the UI and export to py file after that with a command so you can use in your program, if you need any help text me.

[NguyenDucQuan12]

thank you, i will learn about qt

[shine-jayakumar]

how to make exe from python file with paddleOCR

i am using auto-py-to-exe but i think it will work in pyinstaller too with the same parameters to collect data.

install auto-py-to-exe

for installing auto-py-to-exe $ pip install auto-py-to-exe Then to run it $ auto-py-to-exe

github page: https://github.com/brentvollebregt/auto-py-to-exe

select you main file .py and add the paddle dependencies in the Advanced options

it take me a while to discover this. aq é brasil HUE HUE HUE HUE BR

#6875 #9951

Thank you @Heryk13. You solution worked perfectly for me for Python 3.11.
For Python 3.10, I had to edit Python310/Lib/dis.py as per this stackoverflow answer

Thanks a lot again.

[Heryk13]

im glad to hear that. good luck on your project

[wgsthink]

Thank you, your solution solved my problem.

[Heryk13]

😁😁😁 i'm glad to hear that

[tangyanglm]

Thank you, this method also solved my problem

[stephen617hong]

After installing your steps, this prompt appears

[Heryk13]

what python version are you using?
what are the libraries you are using?
what are you trying to do? make a build for exe or installing something?

searching about cublasLt64 maybe is related to something in tensorflow

[stephen617hong]

it's working,thanks.

[Suvoo]

This is a great solution, might miss some corner cases with dll files,
perfectly working solution (August 2024) : #11342

Tl,DR :
pyinstaller main.py --collect-all paddleocr --collect-all pyclipper --collect-all imghdr --collect-all skimage --collect-all imgaug --collect-all scipy.io --collect-all lmdb

[Maximory77]

I tried a build with Pyinstaller with your method and it worked. It created the build without any problem. Unfortunately, when I test the exe created with the build, the execution time is much, much slower. It is so slow that I don't have any output after 30 minutes. When I use my program on runtime python, I start to have ouputs in 10 seconds and the program finish in around 400 s.
For precision, here is the command I used to create my build :

pyinstaller --noconfirm --onedir --console --hidden-import "torch" --hidden-import "pyclipper" --hidden-import "imghdr" --hidden-import "imgaug" --collect-data "paddle" --collect-all "paddleocr" --collect-data "scipy" --collect-all "lmbd" --collect-all "skimage" sliding_window.py

If you have any idea why it is doing it. Please, let me know if you have an idea to fix my problem.

[Heryk13]

sorry no idea what it could be, maybe if you put some print in your code and make the exe, you could log what is happening inside and discover it out from console.

[Maximory77]

Ok now it is working for some reason. I changed my source code and this is how I initialised the ocr function :

slice = {'horizontal_stride': 2000, 'vertical_stride': 1000, 'merge_x_thres': 100, 'merge_y_thres': 60} ocr = PaddleOCR(use_angle_cls=True, gpu_mem = 2000, save_crop_res = False, crop_res_save_dir = output_folder + "/crop", benchmark = False, save_log_path = output_folder, show_log = True, savefile = False, onnx = False, tensorrt=False, lang='latin', use_dilation = True, det_limit_side_len = 2500, max_batch_size = 10, cls_batch_num = 6, rec_batch_num = 6 )

Before I used the same parameters excepts these :
onnx = True, tensorrt = True, gpu_mem = 4000, max_batch_size =20, cls_batch_num = 12, rec_batch_num =12
One or more of these paramaters should be the source of the problem.

[timminator]

I've created a standalone executable version of PaddleOCR that behaves exactly like the command line version of PaddleOCR when it is installed for example via pip.
I've created a CPU and GPU version and also setup installers for both of them for an even easier usage.
If you are interested, you can find my project here.