Augmentation for Object Detection pipeline

[pranavjadhav001]

Going through the repo, I saw mostly image preprocessing and augmentation layers which allows only image transformations. Is there any scope for augmentation transformation of image and bbox for object detection like albumentations library. https://albumentations.ai/docs/getting_started/bounding_boxes_augmentation/
I also see image preprocessing operations have been created as layers to make it more as 1 single coherent model. Not sure, how the support must be added for object detection preprocessing layers where the input is image and labels.

[innat]

@pranavjadhav001 augmentation for obj detection will be added soon. Feel free to contribute.

[pranavjadhav001]

Thanks for the reply. I would love to contribute. Is it planned in the roadmap or will be added later?

[innat]

You can read the roadmap.

[innat]

and design proposal.

[LukeWood]

Thanks for linking this @innat ! I have just checked out this conversation. Yes, we will be adding bbox support as one of our next big ongoing efforts! @divyashreepathihalli is taking on a lot of this, but feel free to pick up layers and add an augment_bounding_box method similar to the augment_image and augment_label calls!

[jbischof]

This is done.