Delete Images in Custom Labels Datasets

Is there a reason removing images is not supported? Removing annotations and leaving unused images in datasets creates noise when managing large datasets. Hoping this feature will be added soon.

Hey Matt

If you remove all label annotations from an image, it will be ignored during the training.

Note that your training might fail in some cases if removing label annotations caused validation rules to fail.
E.g. if you have only 2 labels and you removed all annotations corresponding to 1 label, your training will fail since at least 2 labels are required.

Validation list can be found here (see section "Semantics") - https://docs.aws.amazon.com/rekognition/latest/customlabels-dg/cd-manifest-files-validation-rules.html

Hope this helps!

Thank you, that helps a lot!

If I wanted to remove images that I mistakenly uploaded from a dataset so my model is not trained on them, would you suggest just removing all annotations as they are then essentially ignored from training?

Appreciate the help!

Matt

Matt

Yes you can remove the annotations to avoid them being used for training.

If you have access to the manifest file, you can manually delete the offending rows.

Hey grajamey,
Thank you for your feedback. Deleting images from datasets is on our roadmap. I have also passed on your feedback to the product team to help in the prioritization of this feature.

Hi,

I am unable to delete images from Amazon Rekognition dataset.
Is there any option to do it myself or would you be able to delete it?

Thanks,
Prathyusha

Hi Prathyusha,

Custom labels dataset is managed by a dataset manifest file. If you have access to the manifest file, you can manually delete the offending rows. After deletion you can create a new dataset by importing the updated manifest file.
The manifest file can be accessed through a bucket which name starts with custom-labels-console-*. Within the bucket, you can locate the file with key datasets/<data set name>/manifests/output/output.manifest.
We have an item on our roadmap to support deleting images from dataset and I have also passed on your feedback to the team.

Thanks