无Python环境的PC运行导出可执行文件时出现Numpy错误

在Visual Studio Code中运行程序一切正常；导出为可执行文件后本机运行也没问题，但在另一台无Python环境的Windows 10 22H2电脑上运行时，出现Numpy相关错误：

Traceback (most recent call last):   File "object_detector_from_camera.py", line 85, in <module> TypeError: __init__():
incompatible constructor arguments. The following argument types are supported:
1. mediapipe.python._framework_bindings.image.Image(image_format: mediapipe::ImageFormat_Format, data: numpy.ndarray[numpy.uint8])     
2. mediapipe.python._framework_bindings.image.Image(image_format: mediapipe::ImageFormat_Format, data: numpy.ndarray[numpy.uint16])     
3. mediapipe.python._framework_bindings.image.Image(image_format: mediapipe::ImageFormat_Format, data: numpy.ndarray[numpy.float32])  Invoked with: kwargs: image_format=<ImageFormat.SRGB: 1>, data=None

错误指向代码第85行：image = mp.Image(image_format=mp.ImageFormat.SRGB, data=img)

使用的PyInstaller打包命令：
pyinstaller C:\Users\pc_test1\Desktop\schifo\object_detector_from_camera.py --noconsole --onefile

开发环境版本：

Python 3.12.1
mediapipe                 0.10.14
numpy                     2.0.0
opencv-contrib-python     4.10.0.84
opencv-python             4.10.0.84
pip                       23.2.1
pyinstaller               6.9.0

所有涉及电脑均为Windows 10 22H2系统，怀疑Build文件夹缺失文件，求解决思路。

附上代码：

import os
import cv2
import json
import numpy as np
import time
import sys
import imutils
import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision


# 实现可视化目标检测结果的函数
import cv2
import numpy as np

MARGIN = 10  # 像素
ROW_SIZE = 10  # 像素
FONT_SIZE = 1
FONT_THICKNESS = 1
TEXT_COLOR = (255, 0, 0)  # 红色


def visualize(
    image,
    detection_result
) -> np.ndarray:
  """在输入图像上绘制边界框并返回
  Args:
    image: 输入RGB图像
    detection_result: 需要可视化的所有Detection实体列表
  Returns:
    带有边界框的图像
  """
  for detection in detection_result.detections:
    # 绘制边界框
    bbox = detection.bounding_box
    start_point = bbox.origin_x, bbox.origin_y
    end_point = bbox.origin_x + bbox.width, bbox.origin_y + bbox.height
    cv2.rectangle(image, start_point, end_point, TEXT_COLOR, 3)

    # 绘制标签和置信度
    category = detection.categories[0]
    category_name = category.category_name
    probability = round(category.score, 2)
    result_text = category_name + ' (' + str(probability) + ')'
    text_location = (MARGIN + bbox.origin_x,
                     ROW_SIZE + bbox.origin_y - (MARGIN + 10))
    cv2.putText(image, result_text, text_location, cv2.FONT_HERSHEY_PLAIN,
                FONT_SIZE, TEXT_COLOR, FONT_THICKNESS)

  return image

# 创建ObjectDetector对象
base_options = python.BaseOptions(model_asset_path='efficientdet_lite0.tflite')
options = vision.ObjectDetectorOptions(base_options=base_options,
                                       score_threshold=0.5)
detector = vision.ObjectDetector.create_from_options(options)


pTime = 0
cTime = 0

s = 0
# if len(sys.argv) > 1:
#     s = sys.argv[1]

print(s)
print("aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa")
    

cap = cv2.VideoCapture(s)

while True:
    success, img = cap.read()
    img = cv2.flip(img, 1)
    #imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    #image = mp.Image.create_from_file(img)
    image = mp.Image(image_format=mp.ImageFormat.SRGB, data=img)
    detection_result = detector.detect(image)
    print(detection_result)
    image_copy = np.copy(image.numpy_view())
    annotated_image = visualize(image_copy, detection_result)
    rgb_annotated_image = cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
    #cv2.imshow("temp", rgb_annotated_image)
    cTime = time.time()
    fps = 1/(cTime-pTime)
    pTime = cTime
    annotated_image = imutils.resize(annotated_image, width=720)
    cv2.putText(annotated_image, str(int(fps)), (10,70), cv2.FONT_HERSHEY_COMPLEX, 3, (0,255,255), 3)
    cv2.imshow("window", annotated_image)
    if cv2.waitKey(1) & 0xFF == ord('q'):
          break

• 排查摄像头读取失败问题：错误提示data=None，说明cap.read()返回的img是None，即目标电脑上摄像头读取失败。在代码中加入判断逻辑：

  while True:
      success, img = cap.read()
      if not success or img is None:
          print("无法读取摄像头画面")
          time.sleep(1)
          continue
      img = cv2.flip(img, 1)
      # 后续代码...
  

  同时确认目标电脑是否有摄像头、摄像头是否被占用或需要权限。

• 修复色彩空间不匹配问题：OpenCV读取的图像是BGR格式，而mp.Image的SRGB格式需要RGB数据，必须添加格式转换：

  imgRGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
  image = mp.Image(image_format=mp.ImageFormat.SRGB, data=imgRGB)
  

• 完善PyInstaller打包配置：

  1. 改用--onedir模式打包（去掉--onefile），查看完整依赖文件，对比本机与目标机的运行差异，排查缺失库。
  2. 打包时手动指定隐藏依赖，确保numpy和mediapipe被完整打包：

     pyinstaller C:\Users\pc_test1\Desktop\schifo\object_detector_from_camera.py --noconsole --onefile --hidden-import numpy --hidden-import mediapipe
     

  3. 将efficientdet_lite0.tflite模型文件放到可执行文件同目录，确保程序能加载模型。

• 降级numpy版本：numpy 2.0.0与mediapipe 0.10.14可能存在兼容性问题，降级到1.26.x版本后重新打包：

  pip install numpy==1.26.4
  

内容的提问来源于stack exchange，提问作者Noob_Testautomation