_audio_microfrontend_op.so not found

Question

My Python version is 3.7 (Windows10, 64bit, tensorflow 2.0)

I implemented a program to detect eye-blink using opencv. Press F5 in the pycham to perform normal operation. I made exe file using pyinstaller. However, when I run with the generated exe file, an error occurs.

**............... tensorflow.python.framework.errors_impl.NotFoundError: C:\User\User Name\AppData\Local\Temp\_MEI401122\tensorflow\lite\experimental\microfrontend\python\ops\_audio_microfrontend_op.so not found

[40068] Failed to execute script test**

I've tried numerous ways and found a solution, but it's not working out.

Path on the error message is not on my PC.(Folder _MEI401122 does not exist)

On my PC, the _audio_microfrontend_op.so file is in the path of the picture below.

I have no idea. Please help me.

enter image description here

I add my source code and error content.

My project folder path is as follows.

Path: C:\Users\User Name\Downloads\eye_blink_detector-master

[My Error contents image]

enter image description here

[My source code]

# -*- coding: utf-8 -*-
import cv2, dlib
import numpy as np
from imutils import face_utils
from keras.models import load_model
from time import localtime, strftime
from datetime import datetime
import time
from tkinter import *
import tkinter.messagebox


root = Tk()

WELCOME_MSG = '''Welcome to this event.'''
WELCOME_DURATION = 2000

def welcome():
    top = tkinter.Toplevel()
    top.title('Welcome')
    Message(top, text="카운트", padx=20, pady=20).pack()
    top.after(WELCOME_DURATION, top.destroy)

IMG_SIZE = (34, 26)

#root.mainloop()

detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('shape_predictor_68_face_landmarks.dat')

model = load_model('models/2018_12_17_22_58_35.h5')
model.summary()

count = 0
count_eye_open = 0

f = open("d:/새파일.txt", 'a')


def crop_eye(img, eye_points):
  x1, y1 = np.amin(eye_points, axis=0)
  x2, y2 = np.amax(eye_points, axis=0)
  cx, cy = (x1 + x2) / 2, (y1 + y2) / 2

  w = (x2 - x1) * 1.2
  h = w * IMG_SIZE[1] / IMG_SIZE[0]

  margin_x, margin_y = w / 2, h / 2

  min_x, min_y = int(cx - margin_x), int(cy - margin_y)
  max_x, max_y = int(cx + margin_x), int(cy + margin_y)

  eye_rect = np.rint([min_x, min_y, max_x, max_y]).astype(np.int)

  eye_img = gray[eye_rect[1]:eye_rect[3], eye_rect[0]:eye_rect[2]]

  return eye_img, eye_rect

# main

cap = cv2.VideoCapture(0) #'videos/2.mp4')

while cap.isOpened():
  ret, img_ori = cap.read()

  if not ret:
    break

#윈도우 사이즈
  img_ori = cv2.resize(img_ori, dsize=(0, 0), fx=1.0, fy=1.0)

  img = img_ori.copy()
  gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

  faces = detector(gray)

  dt = datetime.now()

  for face in faces:

    shapes = predictor(gray, face)
    shapes = face_utils.shape_to_np(shapes)

    eye_img_l, eye_rect_l = crop_eye(gray, eye_points=shapes[36:41]) #l_eye_poits = [36, 37, 38, 39, 40, 41] 원소스: [36,42]
    eye_img_r, eye_rect_r = crop_eye(gray, eye_points=shapes[42:47]) #r_eye_points = [42, 43, 44, 45, 46, 47] 원소스: [42:48]

    eye_img_l = cv2.resize(eye_img_l, dsize=IMG_SIZE)
    eye_img_r = cv2.resize(eye_img_r, dsize=IMG_SIZE)
    eye_img_r = cv2.flip(eye_img_r, flipCode=1)

    cv2.imshow('l', eye_img_l)
    cv2.imshow('r', eye_img_r)

    eye_input_l = eye_img_l.copy().reshape((1, IMG_SIZE[1], IMG_SIZE[0], 1)).astype(np.float32) / 255.
    eye_input_r = eye_img_r.copy().reshape((1, IMG_SIZE[1], IMG_SIZE[0], 1)).astype(np.float32) / 255.

    pred_l = model.predict(eye_input_l)
    pred_r = model.predict(eye_input_r)

    # visualize
    state_l = '%.2f' if pred_l > 0.1 else '-%.1f'
    state_r = '%.2f' if pred_r > 0.1 else '-%.1f'

    state_l = state_l % pred_l
    state_r = state_r % pred_r

    # Blink Count
    if pred_l <= 0.1 and pred_r <= 0.1:
        count_eye_open += 1
        print("blinking, "+ str(dt.strftime('%Y-%m-%d %H:%M:%S.%f')))

    cv2.rectangle(img, pt1=tuple(eye_rect_l[0:2]), pt2=tuple(eye_rect_l[2:4]), color=(255,255,255), thickness=2)
    cv2.rectangle(img, pt1=tuple(eye_rect_r[0:2]), pt2=tuple(eye_rect_r[2:4]), color=(255,255,255), thickness=2)

    cv2.putText(img, state_l, tuple(eye_rect_l[0:2]), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255,255,255), 2)
    cv2.putText(img, state_r, tuple(eye_rect_r[0:2]), cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255,255,255), 2)

    cv2.putText(img, "eye blink: " + str(count_eye_open), (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255))
    cv2.putText(img, "Time: " + str(strftime("%S", localtime())), (50, 100), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255))

    if str(strftime("%S", localtime())) == "00":
        count += 1
        if count == 1 and count_eye_open > 0:
            print("Transfer Data...:" + str(count_eye_open))
            f.write("Transfer Data...:" + str(count_eye_open) + "\n")
            count_eye_open = 0
            count = 0
    else:
      count = 0

    time.sleep(0.12)

  cv2.imshow('result', img)
  if cv2.waitKey(1) == ord('q'):
    f.close()
    break

Answer 1

I was able to fix this error by manually specifying to copy the _audio_microfrontend_op.so file inside the spec file.

# -*- mode: python ; coding: utf-8 -*-

import os
import importlib


a = Analysis(
        (...)
             datas=[(os.path.join(os.path.dirname(importlib.import_module('tensorflow').__file__),
                                  "lite/experimental/microfrontend/python/ops/_audio_microfrontend_op.so"),
                     "tensorflow/lite/experimental/microfrontend/python/ops/")],
        (...)
)

Answer 2

For someone who is still confused in spec which is in directory, where is your original filename.py, add this in datas = []

(os.path.join(os.path.dirname(importlib.import_module('tensorflow').__file__),
                              "lite/experimental/microfrontend/python/ops/_audio_microfrontend_op.so"),
                 "tensorflow/lite/experimental/microfrontend/python/ops/")

Then put there import

import os
import importlib

Then run this to compile by the specs you changed.

pyinstaller filename.spec

Answer 3

Don't have enough rep to comment so I'll just post an answer that clarifies things. As Mieszko mentioned in his answer you need to manually specify to copy the _audio_microfrontend_op.so file.

To do so, open the directory that the original filename.py file is in. You'll find a file called filename.spec. Open this file in an editor and add the changes Mieszko specified:

# -*- mode: python ; coding: utf-8 -*-

import os
import importlib


a = Analysis(
    (...)
         datas=[(os.path.join(os.path.dirname(importlib.import_module('tensorflow').__file__),
                              "lite/experimental/microfrontend/python/ops/_audio_microfrontend_op.so"),
                 "tensorflow/lite/experimental/microfrontend/python/ops/")],
    (...)
)

Once you've made those changes, save the file and then run pyinstaller again with the .spec file instead of the .py file

pyinstaller filename.spec