It would be nice to use the albumentations library in Detectron2. The albumentations library provides a wide range of image augmentations and it is therefore useful to integrate with Detectron2, as an addition to the image augmentations that it already provides.

This feature has already been discussed earlier in this issue from 2021, but, as far as I can see, did not lead to a successful merge.

I am building on the code provided by @KUASWoodyLIN, which is a great start. However, for integration with the default DatasetMapper, it is necessary to convert between the Detectron2 bounding box format and the bounding box format expected by albumentations, which I have added. In addition, I also added a test for the new T.Albumentations class. Here's an example of how it can be used:

from detectron2.data import DatasetMapper
import detectron2.data.transforms as T
import albumentations as A

augs = [
    T.Albumentations(A.HorizontalFlip(p=0.5)),
    T.Albumentations(A.RandomBrightnessContrast(p=0.5)),
    T.FixedSizeCrop(crop_size=(256, 256)),
]

mapper = DatasetMapper(cfg, is_train=True, augmentations=augs)

As shown here, the T.Albumentations class can be used in conjunction with other Detectron2 augmentations.

Some limitations of this implementation:

It only works for images, bounding boxes, and masks. Keypoint augmentations could be added relatively easily, by implementing the T.AlbumentationsTransform.apply_coords method.
It does not work for composite transforms, e.g. A.Compose or A.OneOf.

Please let me know if anything is missing or if there are still changes that need to be made.

Hi @tobiasvanderwerff!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks!

facebookresearch / detectron2

Implement wrapper around albumentations #5253

Action Required

Process