Pyinstaller 编译py文件到exe

坐在墙上的猫

已于 2025-08-06 17:32:15 修改

阅读量231

点赞数 1

CC 4.0 BY-SA版权

文章标签：人工智能 python

于 2025-03-07 16:40:25 首次发布

本文链接：https://blog.youkuaiyun.com/ruotianxia/article/details/146096765

Pyinstaller 编译py文件到exe

参考
打包代码
2. 问题2
3. 问题3
4.问题4
5. 问题5

参考

https://zhuanlan.zhihu.com/p/130328237

https://zhuanlan.zhihu.com/p/406790850

https://blog.youkuaiyun.com/moxiao1995071310/article/details/116932406

https://www.jianshu.com/p/38abe3b198af

打包代码

生成spec

pyinstall  **.py  //直接打包
pyinstall -F **.py  //打包成单文件
pyinstall -W **.py  //去掉控制台窗口，黑窗口
pyinstall -i ***.ico **.py  //添加图标 ***

 pyinstaller app.spec

问题1

onnx 版本要和自己cuda和cudnn匹配，否则会找不到相应的dll

Library not found: could not resolve ‘cublas64_12.dll’, dependency of
‘C:\ProgramData\anaconda3\envs\x-anylabeling\lib\site-packages\onnxruntime\capi\onnxruntime_providers_cuda.dll’.

pip install onnxruntime-gpu==1.16.0

2. 问题2

dll load failed while importing onnx_cpp2py_export 动态链接库(DLL)初始化历程失败

解决方法：将System32中的这些dll拷贝到exe目录下：

在这里插入图片描述

导出处spec路径：

pyi-makespec -D main.py （使用：生成多个文件，文件太大，生成一个报错）
pyinstaller –name=your_app_name –onefile –windowed your_script.py
# -*- mode: python ; coding: utf-8 -*-
import sys
from PyInstaller.utils.hooks import copy_metadata

sys.setrecursionlimit(5000)  # required on Windows

a = Analysis(
    ['app.py'],
    pathex=['../code'],
    binaries=[],
    datas=copy_metadata('torchvision'),
    hiddenimports=[],
    hookspath=[],
    hooksconfig={},
    runtime_hooks=[],
    excludes=[],
    noarchive=False,
    optimize=0,
)
pyz = PYZ(a.pure)

exe = EXE(
    pyz,
    a.scripts,
    [],
    exclude_binaries=True,
    name='app',
    debug=True,
    bootloader_ignore_signals=False,
    strip=False,
    upx=True,
    console=True,
    disable_windowed_traceback=False,
    argv_emulation=False,
    target_arch=None,
    codesign_identity=None,
    entitlements_file=None,
)
coll = COLLECT(
    exe,
    a.binaries,
    a.datas,
    strip=False,
    upx=True,
    upx_exclude=[],
    name='app',
)

pyinstaller  app.spec

3. 问题3

serialized_entry = struct.pack( ^^^^^^^^^^^^
struct.error: ‘I’ format requires 0 <= number <= 4294967295
原因是打包成一个exe文件太大了，超出范围，需要拆分成多个文件

   https://blog.youkuaiyun.com/qq_52785898/article/details/127494245

4.问题4

pyinstaller no package metadata was found for torchvision

https://blog.youkuaiyun.com/qq_31679735/article/details/143972630

方法1(实测)：
from PyInstaller.utils.hooks import copy_metadata
a = Analysis(
    ['app.py'],
    pathex=['../code'],
    binaries=[],
    datas=copy_metadata('torchvision'),
    hiddenimports=[],
    hookspath=[],
    hooksconfig={},
    runtime_hooks=[],
    excludes=[],
    noarchive=False,
    optimize=0,
)

方法2：

datas = [
    ('D:/Python/Lib/site-packages/tortoise-orm-0.18.0.dist-info', 'tortoise-orm-0.18.0.dist-info'),
]

5. 问题5

运行程序时，一直循环启动app.exe

暂时未解决

2025-08-01 更新

生成默认 spec 文件

pyinstaller --name="AI标注工具" --windowed main.py

执行后会生成 AI标注工具.spec 文件。

编辑 spec 文件
用文本编辑器打开 AI标注工具.spec，修改以下关键部分：

# -*- mode: python ; coding: utf-8 -*-
# -*- mode: python ; coding: utf-8 -*-
import sys
from pathlib import Path

# 获取项目根目录
root_path = Path(__file__).parent
ultralytics_path = r'D:\yolos\ultralytics\ultralytics'

a = Analysis(
    ['annotation_tool.py'],
    pathex=[str(root_path)],
    binaries=[],
    datas=[],
    hiddenimports=[
        'preimport',
        'ultralytics',
        'ultralytics.*',  # 通配符匹配所有子模块
        'ultralytics.nn.*',
        'ultralytics.yolo.*',
        'ultralytics.data.*',
        'ultralytics.engine.*',
        'ultralytics.utils.*',
        'PyQt5.QtXml',  # 可能缺失的PyQt5子模块
    ],
    datas=[(ultralytics_path, 'ultralytics')],  # 格式：(源路径, 打包后路径)
    hookspath=[],
    hooksconfig={},
    runtime_hooks=[],
    excludes=[],
    win_no_prefer_redirects=False,
    win_private_assemblies=False,
    cipher=None,
    noarchive=False,
)
icon_files = [(str(f), str(f.parent.relative_to(root_path)))
             for f in root_path.glob("App/Packages/icons/*") if f.is_file()]

pyz = PYZ(a.pure, a.zipped_data, cipher=None)

exe = EXE(
    pyz,
    a.scripts,
    a.binaries,
    a.zipfiles,
    a.datas + icon_files,  # 添加资源文件
    name='AI标注工具',
    debug=False,
    bootloader_ignore_signals=False,
    strip=False,
    upx=True,  # 压缩EXE体积
    upx_exclude=[],
    runtime_tmpdir=None,
    console=True,  # 隐藏控制台
    disable_windowed_traceback=False,
    argv_emulation=False,
    target_arch=None,
    codesign_identity=None,
    entitlements_file=None,
    icon=str(root_path / "App/Packages/icons/airbnb.png"),  # 图标路径
)
coll = COLLECT(
    exe,
    a.binaries,
    a.datas,
    strip=False,
    upx=True,
    upx_exclude=[],
    name='SharetekAI',
)

)

pyinstaller AI标注工具.spec --clean  # 清除缓存，确保新配置生效

2025-08-04

问题1：DLL load failed while importing onnx_cpp2py_export
动态链接库(DLL)初始化例程失败

# 卸载现有库
pip uninstall onnx onnxruntime

# 重新安装（指定较稳定的版本）
pip install onnx==1.16.1 onnxruntime-gpu==1.16.1   onnxruntime==1.6.1(这三个都要安装相同的版本)

如果安装后matplotlib debug出现问题，重新安装一下
pip install matplotlib

bash 问题2：DLL load failed while importing tensorrt : 找不到指定的模块

pip install tensorrt-8.6.1-cp311-none-win_amd64.whl

将tensorrt的libs中的库拷贝过去（后面重新编译后，不加也能够正常启动）

在这里插入图片描述

.spec 1

# -*- mode: python ; coding: utf-8 -*-
import sys
from pathlib import Path
import os
import onnx
import onnxruntime
import tensorrt

# 获取项目根目录
root_path = os.getcwd()
import ultralytics
ultralytics_path = os.path.dirname(os.path.dirname(ultralytics.__file__))

# 获取onnx和onnxruntime的安装路径
onnx_path = os.path.dirname(onnx.__file__)
#onnxruntime_path = os.path.dirname(onnxruntime.__file__)
tensorrt_path = os.path.dirname(tensorrt.__file__)


a = Analysis(
    ['annotation_tool.py'],
    pathex=[os.path.join(str(root_path), "App")],
    binaries=[],
    hiddenimports=[
        'preimport',
        'ultralytics',
        'ultralytics.yolo',
        'ultralytics.nn',
        'onnx',
        'onnxruntime',
        'onnxruntime.capi',
        'onnxruntime.capi._pybind_state',
        'onnxruntime.quantization',
        'onnxruntime.tools',
    ],
    datas=[
        (ultralytics_path, 'ultralytics'),
        (onnx_path, 'onnx'),
        #(onnxruntime_path, 'onnxruntime'),
        (tensorrt_path, 'tensorrt'),
        # 如果你有其他需要打包的数据文件，可以在这里添加
    ],
    hookspath=[],
    hooksconfig={},
    runtime_hooks=[],
    excludes=[],
    win_no_prefer_redirects=False,
    win_private_assemblies=False,
    cipher=None,
    noarchive=False,  # 确保所有模块都被归档，不生成单独的.pyz文件
)

pyz = PYZ(a.pure)

exe = EXE(
    pyz,
    a.scripts,
    [],
    exclude_binaries=True,
    name='AI 2.0.0',
    debug=False,
    bootloader_ignore_signals=False,
    strip=False,
    upx=True,
    console=True,
    disable_windowed_traceback=False,
    argv_emulation=False,
    target_arch=None,
    codesign_identity=None,
    entitlements_file=None,
    icon='resources/images/icon.icns'
)

coll = COLLECT(
    exe,
    a.binaries,
    a.datas,
    strip=False,
    upx=True,
    upx_exclude=[],
    name='AI 2.0.0',
)