Dataset for benchmarking Multiple Object Tracking and Segmentation (MOTS) in an apple orchard field.

A dataset of temporally consistent apple images and labels taken using UAVs and a wearable sensor in an orchard, consisting of 86000 manually annotated apple instances and 1700 frames annotated in the MOTS (Multi-object Tracking and Segmentation) style.

Sequence 0-5 are used for training. Sequence 6-8 are used for testing/validation. Sequence 10-12 are the testing datasets that have "ignore regions" overlays.

The code used in the paper can be found on our GitLab.