Open latoa opened 3 years ago
Hi, maybe duplicating is not the best decription, since augmented image would be transformed and normally not the same as the original image. The 'imgaug' library was used. Please find more information at https://imgaug.readthedocs.io/en/latest/index.html. The image is augmented once. As for how the augmentation was applied, please check out the code in the original Mask_RCNN page. https://github.com/matterport/Mask_RCNN/blob/master/mrcnn/model.py
Could you please confirm that data augmentation is duplicating the train and validation sets? Each image is only transformed once, right?