keras same image augmentation to both X and Y

[yarin0...@gmail.com]

Hello everybody,
The problem i wanna solve include image segmentation. Therefore, the same augmentation that randomed for an original image must be the same for the image-target.

For example, when generating new instance of image with vertical-flip the same operation should be done for the target image which consists with the true segmentation pixel-wise.

Is there an easy way to make that with keras?

Much thanks!

[james.g...@gmail.com]

Keras does have the functionality to do what you are looking for.

https://keras.io/preprocessing/image/

You can have an imagedatagenerator for both the original image and the segmented mask.

If you go that way, I would suggest writing the files to disk in a test run to confirm that you are doing what you think you are doing.

I found that it is best it to write your own generator and have complete control.

You can look in to the keras source code for an example.

[Joe]

Have a look at https://github.com/aleju/imgaug