FACE COUNTING (count the number of faces present in an image)¶

!pip install mtcnn

# Loading required libraries
from PIL import Image
from numpy import asarray
from mtcnn.mtcnn import MTCNN 
import tensorflow.keras as Keras
import matplotlib.pyplot as plt
import os
import pandas as pd
import numpy as np

# CSV file containing image name as an input and number of faces as an output
dataset= pd.read_csv('.../face_count.csv')

dataset.head()

print('Dataset Shape : {}'.format(dataset.shape))

# image directory
filename = '.../image_data'

#loading sample image
dir = filename + '/' + dataset['Name'][0]  # loading first image of csv file
image = Image.open(dir)                    #loading image from given path
image = image.convert('RGB')               # converting image into 3D (RGB) image
pixels = asarray(image)

plt.imshow(pixels)
plt.axis('off')
print('Number of Faces : {}'.format( dataset['HeadCount'][0]))

#face extraction using MTCNN
detector = MTCNN()     
results = detector.detect_faces(pixels)

# As we already know that there are 4 faces so results variable contains  %u2018box%u2019, %u2018confidence%u2019 
# and %u2018keypoints%u2019of each face
total_faces = len(results)

for face in range(total_faces):
  print('box : {}'.format(results[face]['box']))
  print('confidence : {}'.format(results[face]['confidence']))
  print('keypoints : {}'.format(results[face]['keypoints']))

# similarly now applying over 1000 images to measure performance of MTCNN
# storing number of heads in each and every image
output = []

# looping through CSV file
for img in dataset['Name']:
    
    # diretory of image
    dir = filename + '/' + img
    image = Image.open(dir)            #loading image from given path
    image = image.convert('RGB')       # converting image into 3D (RGB) image
    pixels = asarray(image)            
    
    # face extraction
    detector = MTCNN()                 
    results = detector.detect_faces(pixels)

    # len(results) returns number of faces in an image
    output.append(len(results))    

# converting output list into dataframe 
output = np.array(output)
output = pd.DataFrame(output)
output.columns = ['Predicted']
output.head()

# actual output 
actual = pd.DataFrame(dataset['HeadCount'])

actual.head()

print('output shape : {}'.format(output.shape))

# Now we have to compare actual output with the predicted output
# accuracy = (predicted[i] == actual[i]) / number_of_images

def get_acc(predicted, actual, num_of_images):
  
  correct = 0
  for idx in range(num_of_images):
    if output['Predicted'][idx] == actual['HeadCount'][idx]:
      correct += 1

  accuracy = correct / num_of_images
  return accuracy

num_of_images = output.shape[0]   # total number of images

accuracy = get_acc(output, actual, num_of_images)
print('accuracy : {}'.format(accuracy))