We select a new dataset for segmentation based on high image variety

[andandandand]

We check for near duplicates between train, test, and validation using the leaky splits functionality of FiftyOne

https://docs.voxel51.com/brain.html#leaky-splits

We use FiftyOne's uniqueness score to pick images with low chance of being near duplicates.

https://docs.voxel51.com/api/fiftyone.brain.internal.core.representativeness.html

The representativeness score can be used to select from these images.

https://docs.voxel51.com/api/fiftyone.brain.internal.core.representativeness.html