Add plots and additional metrics

[ORippler]

Is your feature request related to a problem? Please describe. Imo, it would be nice to have:

• Pixel-wise AUPR + PRO until 30% FPR as metrics
• Plots with baselines where applicable for all metrics (ROC-Curve, PR-Curve, PRO-Curve)

Describe the solution you'd like

1. Add/implement appropriate torchmetric classes, and add them to the AnomlibMetricCollection
2. Add callbacks which access the underlying scores for metric calculation (i.e. tuple of tpr/fpr for auroc) and which subsequently use the scores to generate and save the plots.
   • Alternatively, one could think of expanding/inheriting the VisualizerCallback

Thoughts?

[ashwinvaidya17]

@ORippler I think segregating ImageVisualizerCallback and MetricVisualizerCallback might be a cleaner approach. Let me know what are your thoughts on that.

Also, just to break down your issue here is a list to capture it. Feel free to suggest any modifications.

• [ ] Add AUPR metrics
• [ ] Add PRO metrics
• [ ] Add visualizer for plotting the underlying scores in each metrics

[ashwinvaidya17]

Related issues

• https://github.com/openvinotoolkit/anomalib/issues/321

[ORippler]

@ORippler I think segregating ImageVisualizerCallback and MetricVisualizerCallback might be a cleaner approach. Let me know what are your thoughts on that.

Also, just to break down your issue here is a list to capture it. Feel free to suggest any modifications.

* [ ]  Add AUPR metrics

* [ ]  Add PRO metrics

* [ ]  Add visualizer for plotting the underlying scores in each metrics

Sure we can split up the Callbacks, but then I suggest inheriting from a VisualizerCallbackBase.

[samet-akcay]

Sure we can split up the Callbacks, but then I suggest inheriting from a VisualizerCallbackBase

Sure, this separation should cause minimal duplication.