Module 08 Summary

What this module established

Four numbers, four conventions, and no error on mismatch. Convert at the boundary and assert. Report the IoU threshold with every detection metric - it moves every number without changing the model.

Carry forward

• Non-maximum suppression turns repeated detections into objects. Its threshold only matters on crowded images, so tune it there.
• Precision and recall trade against each other at every threshold. Report which threshold produced the pair.
• Compute your smallest target's size after the resize. Under a few pixels, tiling is the answer and a larger model is not.

Before moving on

Move on when your box format is asserted at the boundary, your IoU threshold is reported with every metric, and you have measured your smallest object after the resize.

Mark complete