Is it possible the directoryiterator or flow_from_directory takes multiple directories as input ? about keras-preprocessing HOT 4 CLOSED

This is a really particular usecase.
I think you should use flow_from_dataframe to solve your issue.

from keras-preprocessing.

Yes I can put all my data to dataframe however then flow_from_dataframe needs the directory also as an argument. If I could give a directory I would use the flow_from_directory. I don't understand why flow from dataframe needs directory.
I created my dataframe as here:
for train_region in training_regions:
X = numpy array from train_region
pd_dataset = pd.DataFrame({'label': Y, 'images': list(X)}, columns=['label', 'images'])
pd_datasets.append(pd_dataset)
data_train = pd.concat(pd_datasets)
The problem is I can not have one directory for all training images ...

from keras-preprocessing.

#56
Don't understand either why it needs the directory argument and x_col needs to be filenames only (not full path) otherwise it raises error

from keras-preprocessing.

This has been solved in the current version. This issue can be closed @Dref360

from keras-preprocessing.