ferdata_to_csv.py

import argparse 
import cv2
import numpy as np
import os
import pandas as pd


def parse_args():
    ''' Argument Parser. '''
    parser = argparse.ArgumentParser()
    parser.add_argument('--data_dir', type=str, required=True, \
        help='Data directory path containing folders for persons'
            ', where each person folder in turn consists of emotion folders.')
    args = parser.parse_args()
    return args

if __name__ == '__main__':
    args = parse_args()
    if not os.path.exists(args.data_dir):
        raise ValueError('Please pass a valid data directory.')

    dirs = os.listdir(args.data_dir)
    # iterate over all persons 
    for person in dirs:
        person_dir = os.path.join(args.data_dir, person)
        # skip the file if it is not a directory
        if not os.path.isdir(person_dir):
            continue 
        person_csv = os.path.join(args.data_dir, '%s.csv' %(person))

        # open csv file for person
        with open(person_csv, 'w') as f:
            f.write('emotion,pixels,usage\n')
            # iterate over emotion folders
            for emotion in os.listdir(person_dir):
                emotion_label = int(emotion)
                emotion_dir = os.path.join(person_dir, emotion)
                # iterate over images in the emotion folder
                for image_name in os.listdir(emotion_dir):
                    image = cv2.imread(os.path.join(emotion_dir, image_name), cv2.IMREAD_GRAYSCALE)
                    image = image.reshape(-1)
                    image_str = [str(pix) for pix in image] # convert image pixels to str
                    image_str = " ".join(image_str)
                    f.write('%d,%s,%s' %(emotion_label, image_str, 'training'))
                    f.write('\n')

        print ('completed writing csv for %s' %(person))
    
    print ('completed process')

    # Reading the last csv file
    pdata = pd.read_csv(person_csv)
    print (pdata.head())
    # converting the pixels from string to np array
    pdata['pixels'] = pdata['pixels'].apply(lambda image_px : np.fromstring(image_px, sep = ' ')) 
    images = pdata.iloc[:, 1].values
    images = np.vstack(images).reshape((-1, 48, 48, 1)).astype('float32') # images present in the csv file
    print ('images', images.shape)