Metrics

Challenge pages point here for metric definitions and rationale.

Adjusted Rand Index

Adjusted Rand Index measures agreement between two partitions while correcting for chance.

In compact form, $ARI$ compares an observed rand index against its expected value under chance:

$$ARI = \frac{RI - \mathbb{E}[RI]}{\max(RI) - \mathbb{E}[RI]}$$

Panoptic Quality

Panoptic quality combines recognition quality and segmentation quality for instance-aware segmentation tasks.

Masked-hit Recovery

Masked-hit recovery measures whether a method reconstructs intentionally hidden detector activity from the visible event context.

Linear Probe Suite

Linear probe suite scores evaluate whether a frozen representation supports held-out detector tasks through lightweight supervised probes.