OoDIS: Anomaly Instance Segmentation Benchmark

Anomaly Instance Segmentation is a task that aims to find objects that are present only at inference time and unknown during training. A typical anomaly is a deer or a cardboard box in the middle of the road. In the benchmark, we extend the labels of well-known benchmarks such as SegemntMeIfYouCan and FishyScapes Lost and Found to instance segmentation. We combine two benchmarks into a unified benchmark and evaluate the most common metrics instance metrics of Average Precision.

We extend labels of 2 existing benchmarks for Instance Segmentation:

• SegmentMeIfYouCan, containing two datasets: RoadAnomaly21 and RoadObstacle21
• Fishyscapes: Lost & Found