ID vs OOD Score Distributions

Two overlapping kernel density curves showing ID/OOD separability with mean markers and threshold line.

Prompt

A kernel density plot showing two overlapping distributions of an anomaly score R.

X-axis: Score R, range 0 to 5, gridlines every 1.
Y-axis: Density, range 0 to 1.5.

Two distributions:
- ID (in-distribution): blue curve, mean 0.7, std 0.4. Filled at 25% opacity.
- OOD (out-of-distribution / hazard): coral curve, mean 2.5, std 0.8. Filled at 25% opacity.

Annotations:
- Vertical dashed line at the chosen threshold R = 1.5 with label "Threshold (95th percentile of ID)".
- Vertical solid lines at each distribution's mean, labeled "ID mean" and "OOD mean".
- Top-right legend with the two distributions and an additional line showing the AUROC value (e.g., AUROC = 0.96).

A small inset on the bottom-right shows ROC for the same scores.

Style: clean academic chart, white background, sans-serif labels, restrained palette. Suitable for anomaly detection figures and safety evaluation reports.

Use in Generator

When to use

For OOD detection / anomaly detection / safety-evaluation papers.

Variations

Per-hazard-type breakdown

Replace the single OOD curve with three OOD curves (one per hazard type) and re-color them. Compute AUROC per hazard type and list each in the legend.

Tips

Use translucent fills with thicker borders. Solid fills hide the overlap region.
Annotate threshold and mean lines. Without them the plot is just two blobs.
Show AUROC explicitly somewhere on the plot — readers expect it for binary separability.

FAQ

How do I show shifted distributions over time?

Animate via small multiples: 4 panels in a row, each showing the OOD distribution at a different time step, with the threshold line held constant.