Mask_rcnn: How to use standard data augmentation?

[Paulito-7]

Hello everyone,

I saw that in 'load_image_gt' it is possible to add an option for data augmentation and that this one is used in 'data_generator` used in 'inspect_data.ipynb'. Yet in the notebook I can't see where you can use it for data augmentation for training.
It seems that it is automatically called for training, is that right?

Thanks!

[waleedka]

It used to be automatically called for training. And it did right/left flipping 50% of the time.

But as of earlier today, I pushed an update that makes two changes:

1. Augmentation is not done automatically anymore. Instead you need to pass the augmentation parameter to train().
2. The model now uses the imgaug library for augmentation, and that allows a very wide range of possible augmentations.

The commit is here: https://github.com/matterport/Mask_RCNN/commit/c7e3aa7a32a2ff4e265ef9746ca87dd91ed6bfa2

[Paulito-7]

Thank you for your answer!
I didn't see the commit, it looks great!