Day 77: The calibration problem: why pixels ≠ cm
The honest hard part
Everything so far produces measurements in *pixels*. The size recommendation needs *centimeters*. The conversion is fundamentally ambiguous: the same real-world object appears larger or smaller depending on its distance from the camera and the lens. A 300-pixel shoulder could be a small person up close or a large person far away — you cannot tell from pixels alone. This is the calibration problem, and the roadmap is explicit that it's the honest hard part of Stage 2.
Don't pretend this away
The tempting move is to hand-wave calibration and quote impressive-looking centimeter accuracy. The roadmap forbids it: measured error, not claimed accuracy, is what separates this from a tutorial portfolio. You need a real reference to resolve the scale ambiguity — and you need to be honest about the error bars it leaves. That honesty is the deliverable, as much as the number.
Key terms
- Calibration
- Establishing the pixels-to-real-units (cm) conversion, which requires a reference of known real-world size.
- Scale ambiguity
- The fact that an object's pixel size depends on distance and lens, so pixels alone cannot determine real size.
Why can't you convert a pixel measurement directly to centimeters without a reference?
