How does HeadShape AI measure my baby's head shape?

What happens between my photo and the scores?

1. 1

   A standardised top-down photo. You photograph your baby's head from directly above, with the top of the head pointing to the top of the frame. That framing isn't a style choice — the section below on photo angle explains why it's load-bearing for the measurement.

2. 2

   The outline is traced automatically. An automatic model finds the head in the photo and traces its outline — no manual clicking or cropping needed.

3. 3

   Two proportions are read off that outline. A width-to-length ratio (how wide the head reads relative to how long it reads) and a left-right difference (how closely one side mirrors the other).

4. 4

   Those two proportions become three 0-100 scores. A Proportion Score, a Symmetry Score, and an overall Harmony Score that combines both. The formulas for each are below.

What is the Proportion Score, exactly?

Width is the horizontal span across the traced outline; length is the vertical span. Divide one by the other and you get the ratio the Proportion Score is built on.

Proportion Score = 100 − 200 × |ratio − 0.80|

Full marks — 100 — land exactly at a width-to-length ratio of 0.80. Move away from 0.80 in either direction and the score comes down by the same two points for every 0.01 the ratio moves: a head that reads a little wider than 0.80 loses exactly as much as one that reads the same distance longer than 0.80. Neither direction is treated as the safer one.

Why is 0.80 the target ratio?

0.80 sits at the midpoint of the typical range for this ratio. It's also not a new number picked for this rewrite — it's the same target ratio the earlier version of this tool used, so a Proportion Score means the same thing today as it did before.

What is the Symmetry Score, exactly?

Two diagonal measurements are taken across the outline, one 30° to the left of vertical and one 30° to the right. The closer those two diagonals are to each other, the more symmetric the outline reads.

Symmetry Score = 100 − 650 × max(0, difference − 0.5%)

"Difference" is the gap between the two diagonals, as a share of the longer one. Every real head has some small left-right difference, so the first 0.5% of that gap costs nothing — deductions only start once the difference passes that band, and then they add up quickly.

What is the Harmony Score?

The Harmony Score is the geometric mean of the Proportion and Symmetry scores — the square root of their product, not a plain average.

Harmony Score = √(Proportion Score × Symmetry Score)

That choice matters: a geometric mean punishes a weak score harder than a plain average would, so a strong score on one side can't fully cover for a weak score on the other. A head that scores 100 on one and 40 on the other lands at a Harmony Score of about 63 — noticeably below the 70 a plain average would give it.

Why does the photo have to be top-down, with the top of the head pointing up?

The Symmetry Score depends on comparing two diagonals against a straight reference line, and that reference line is simply "up" in your photo — the vertical edge of the image itself. That's why the framing matters: look straight down, keep the top of the head pointing to the top of the frame, and the reference line is exactly where the measurement needs it to be.

There's a reason it's built this way rather than the more obvious-sounding alternative — letting the outline set its own reference line, for instance by running it through the outline's own longest axis. That approach fails on exactly the heads it's supposed to measure: if a head is asymmetric, the outline's own long axis tilts along with that asymmetry. Measure symmetry against a reference that has already tilted to match the asymmetry, and the asymmetry mostly cancels itself out of the result — the measurement would end up flattering the very heads it's meant to describe accurately. Using the photo's own vertical instead — a line fixed by how you hold the camera, not one the outline gets to choose for itself — keeps the reference honest.

What happens if the outline doesn't look right?

Every traced outline is automatically compared against an ellipse fitted to its own shape. A clean head outline sits close to that fitted ellipse; an outline that has picked up the edge of a blanket, a swaddle, or nearby bedding along with the head departs from it in ways a clean outline doesn't. When that gap is too large, that result is flagged for review (recorded, not yet used to hold back a result). See the validation page for exactly how well that check works and what it costs.

Formulas at a glance