keras flow_from_dataframe for image segmentation (i.e. with a binary mask as label)

Question

Is there a way to use the method "flow_from_dataframe" from keras ( https://keras.io/api/preprocessing/image/#flowfromdataframe-method ) to load an image with a mask? Such that instead of y being a category, y is a path such that a binary mask gets loaded together with the image.

An example dataframe to load the data would be:

+---+---------------------+---------+-------------------------+
|   | img_path            | subject | mask_path               |
+---+---------------------+---------+-------------------------+
| 0 | images/DSC_4540.TIF | 4540    | masks/DSC_4540_mask.tif |
+---+---------------------+---------+-------------------------+
| 1 | images/DSC_9228.TIF | 9228    | masks/DSC_9228_mask.tif |
+---+---------------------+---------+-------------------------+
| 2 | images/DSC_6674.TIF | 6674    | masks/DSC_6674_mask.tif |
+---+---------------------+---------+-------------------------+
| 3 | images/DSC_3453.TIF | 3453    | masks/DSC_3453_mask.tif |
+---+---------------------+---------+-------------------------+
| 4 | images/DSC_6808.TIF | 6808    | masks/DSC_6808_mask.tif |
+---+---------------------+---------+-------------------------+
| 5 | images/DSC_5494.TIF | 5494    | masks/DSC_5494_mask.tif |
+---+---------------------+---------+-------------------------+

Answer 1

You can create a separate ImageDataGenerator for the masks and then zip it with the existing ImageDataGenerator.

Check out the third example from the documentation .