create_gender_neutral_data.py 文件源码

def read_fold(fold_csv_prefix,fold_image_prefix,fold_names):
    width = 256
    height = 256
    i=0

    for fold in fold_names:
        print("Reading fold: %s" % fold)
        df = pd.read_csv(fold_csv_prefix+fold+'.csv')
        inputimages = []
        genders = []
        ages = []


        for index, row in df.iterrows():
            yaw_angle = row['fiducial_yaw_angle']
            gender = row['gender']
            age = row['age']

            if ((gender!='u') and (gender!='Nan') and (age!='None') and (gender!=' ') and (age!=' ') and (yaw_angle >= -45) and (yaw_angle <= 45)):
                    folder_name = row['user_id']
                    image_name = row['original_image']
                    face_id = row['face_id']

                    age_tuple = make_tuple(age)
                    age_id = get_age_range_id(age_tuple)

                    image_path = fold_image_prefix+folder_name+'/landmark_aligned_face.'+str(face_id)+'.'+image_name
                    image = Image.open(image_path)

                    #Resize image
                    image = image.resize((width, height), PIL.Image.ANTIALIAS)
                    image_arr = np.array(image)
                    #image_arr = exposure.equalize_hist(image_arr)


                    if(gender == 'm'):
                        g=0
                    else:
                        g=1

                    inputimages.append(image_arr)
                    genders.append(g)
                    ages.append(age_id)

        print('Done: {0}/{1} folds'.format(i, len(fold_names)))
        i=i+1

        print ('Fold Name: %s' % fold)            
        print ('Images: %i, Gender: %i, Ages: %i' % (len(inputimages), len(genders), len(ages)))            
        print ('')

        currDict = {'fold_name': fold, 'images': inputimages, 'genders': genders, 'ages': ages}
        save_pickle(currDict,fold, '/home/narita/Documents/pythonworkspace/data-science-practicum/gender-age-classification/gender_neutral_data/')