flow_dataframe.py

import numpy as np
import pandas as pd
import os

def flow_from_dataframe(img_data_gen, in_df, path_col, y_col, **dflow_args):
    """
    Creates a DirectoryIterator from in_df at path_col with image preprocessing defined by img_data_gen. The labels
    are specified by y_col.

    :param img_data_gen: an ImageDataGenerator
    :param in_df: a DataFrame with images
    :param path_col: name of column in in_df for path
    :param y_col: name of column in in_df for y values/labels
    :param dflow_args: additional arguments to flow_from_directory
    :return: df_gen (keras.preprocessing.image.DirectoryIterator)
    """
    print('flow_from_dataframe() -->')
    base_dir = os.path.dirname(in_df[path_col].values[0])
    print('## Ignore next message from keras, values are replaced anyways')
    # flow_from_directory: Takes the path to a directory, and generates batches of augmented/normalized data.
    # sparse: a 1D integer label array is returned
    df_gen = img_data_gen.flow_from_directory(base_dir, class_mode='sparse', **dflow_args)
    # df_gen: A DirectoryIterator yielding tuples of (x, y) where x is a numpy array containing a batch of images
    # with shape (batch_size, *target_size, channels) and y is a numpy array of corresponding labels.
    df_gen.filenames = in_df[path_col].values
    df_gen.classes = np.stack(in_df[y_col].values)
    df_gen.samples = in_df.shape[0]
    df_gen.n = in_df.shape[0]
    df_gen._set_index_array()
    df_gen.directory = base_dir  # since we have the full path
    print('Reinserting dataframe: {} images'.format(in_df.shape[0]))
    print('flow_from_dataframe() <--')
    return df_gen