This class is used to extract images from various directories

data_extraction.py

REBUILD_DATA = True

if REBUILD_DATA: #if we are running it for the first time
    data_path = Path('F:/FILES/AI/face-mask-dataset/')
    maskPath = data_path/'dataset1/AFDB_masked_face_dataset'
    maskPath2= data_path/'dataset2/webface_masked'
    nonMaskPath = data_path/'dataset1/AFDB_face_dataset'
    path_dirs = [ [maskPath,1],[nonMaskPath,0] ] #path and label
    if not os.path.exists(data_path):
        raise Exception("The data path doesn't exist")
        
class MaskvNoMask():
    IMG_SIZE = 100
    LABELS = {'NON_MASKED': 0, 'MASKED': 1}
    training_data = [] # We will append one image at a time
    
    def make_training_data(self):
        for data_dir, label in path_dirs:
            print('Reading from: ',label)
            for folder in tqdm(os.listdir(data_dir)):
                folder_path = os.path.join(data_dir, folder)
                for imgpath in os.listdir(folder_path):
                    self.count += 1                       
                    img_path = os.path.join(folder_path, imgpath)
                    try: # putting this in try block will help to avoid corrupt image files
                        img = cv2.imread(img_path)
                        img = cv2.resize(img, (self.IMG_SIZE,self.IMG_SIZE))
                        self.training_data.append([np.array(img), label]) # adding image to train data
                        
                        if label == 1:
                            self.LABELS['MASKED'] += 1 # counting number of masked images
                        if label == 0:
                            self.LABELS['NON_MASKED'] +=1
                            
                    except:
#                         raise Exception('error: {}'.format(img_path))
                        pass
            print(self.LABELS)                        
        
        np.random.shuffle(self.training_data)
        np.save("./npy/training_data.npy", self.training_data) # saving the numpy array so we dont have to read it again, its a lengthy process to read images from disk.
        
if REBUILD_DATA:
    maskvnomask = MaskvNoMask()
    maskvnomask.make_training_data()
    training_data = maskvnomask.training_data
else:
    training_data = np.load('./npy/training_data.npy', allow_pickle=True)