如何使用包含文件名和标签的数据帧将带有类标签的图像从单个目录加载到Keras模型中？

Question

I have two separate folders "train_images" and "val_images". I have all the images inside these two folders. I am doing a categorical classification using DenseNet121 model. I have two different y_df (dataframe) for training and validation. I know about flow_from_directory but for that I need to create different subforlders inside each "train_images" and "val_images" based on classes.

I am willing to to know if I don't want to create those subforlders and don't use flow_from_directory , how do I pass those images into the model? I am using Keras with Tensorflow as the backend.

Answer 1

If you have a dataframe that contains the image filenames along with their labels, you can use the newly introduced flow_from_dataframe() method which has been exactly defined for this purpose. It takes a Pandas dataframe which contains the filenames of the images in one column and their corresponding labels in another column(s). It has not been officially released yet, so you need use the latest development version of Keras on the Keras Github repository . It is expected to be in the next release, though.

Update: It has been included in Keras Preprocessing 1.0.4 and Keras 2.2.3 .

Also, there is a tutorial about this method as well.