Error generating flip and 90deg augmentations for segmentation masks?

I’m trying to create an image data set labelled with segmentation masks.

On attempting to create flips and 90 degree rotations as augmentations, the masks in the rotated images are shrunk and do not orient with the objects identified on the labels.

Attached are some screenshots that show a difference on my screen.

Augmentations

Images in data set:


Some images have distorted label masks. I notice that these are the only images taken with an iphone.

All other images generated with augmentations correctly are from android phones.

Should I assume that these are just display errors and that the labels are assigned correctly?

I appreciate your help very much. Thanks :pray:

Based on the information provided in the extracts, it’s not entirely clear why the masks in your rotated images are shrinking and not orienting correctly with the objects identified on the labels. However, there are a few things you could consider:

  1. Image Augmentation: Roboflow supports various image augmentations including flips and 90-degree rotations. These augmentations are applied to existing images in your dataset to improve the ability of your model to generalize and thus perform more effectively on unseen images. However, it’s important to use these augmentations judiciously. As mentioned in the Roboflow blog, having too many augmented images can harm the performance of a model. Therefore, it’s recommended to stick to realistic scenarios when using augmentations.

  2. Image Quality: The quality of the images used in your dataset could also be a factor. If the images taken with an iPhone are of a different quality or format than those taken with Android phones, this could potentially affect the augmentation process and the resulting masks.

  3. Segmentation Masks: A segmentation mask is a specific portion of an image that is isolated from the rest of an image. You can use the output of a segmentation mask to copy exact areas of an image that have been assigned a label in a computer vision model. If the masks are not aligning correctly with the objects, it could be an issue with how the masks were created or applied.

Without more specific information or the ability to directly examine your dataset and augmentation process, it’s difficult to provide a definitive answer. It’s not recommended to assume these are just display errors. It would be best to investigate further to ensure the labels are assigned correctly.

For more detailed assistance, you might want to refer to the Roboflow documentation or reach out to their support team.


Hi! I’m Roboflow’s AI powered bot. This is an automated response powered using AI with knowledge from our docs, blog, knowledge base, and previous forum responses. If you find this to be a helpful answer, please mark it as the solution using the checkbox icon below. If this doesn’t solve your issue, please follow up with what you’ve tried/why, and the community will continue to chime in as usual.

Thanks. By removing the pre-processing step Auto-Orient, I could add augmentations without any problems with the masks being distorted.

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.