You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I took a look at writing custom augmenters.
It seems both _augment_images() and _augment_keypoints have access to images and keypoints separately.
So I can't read a keypoint(bounding box) and perform augmentation on the image(let's just say I want to assign gray pixels to the entire bounding box area).
Is there any recommended way to approach this problem?
The text was updated successfully, but these errors were encountered:
This is currently a shortcoming of the library. Augmentations where one type of input data (images, heatmaps, segmentation maps, keypoints, bounding boxes, polygons, line strings) requires information of another type are not possible. It is planned to be changed at some point by giving more control to augment(), but right now such augmentations have to be handled outside of the library.
For future users finding the issue: This is now possible since PR #556 using BlendAlphaBoundingBoxes(foreground=child_augmenters). Currently only in master, but will be in the next release (0.4.0). Requires to augment both images and bounding boxes in a single function call, e.g. using BlendAlphaBoundingBoxes(...)(images=<array>, bounding_boxes=<bbs>).
I took a look at writing custom augmenters.
It seems both
_augment_images()
and_augment_keypoints
have access to images and keypoints separately.So I can't read a keypoint(bounding box) and perform augmentation on the image(let's just say I want to assign gray pixels to the entire bounding box area).
Is there any recommended way to approach this problem?
The text was updated successfully, but these errors were encountered: