1 Print the marker sheet
Print at 100% / "Actual size", never "Fit to page". Then check the 100 mm line on the sheet with a ruler.
2 Take the photo
- Sheet on a flat table, in soft even light. No flash, no hard shadows.
- One ring, lying flat inside the corner marks, with white paper showing through it.
- Phone flat and directly above the ring, about 30 cm away. No zoom.
- Marker and ring both fully in frame. Tap to focus, hold still.
3 Upload it
Loading the measurement engine…
The photo is measured on your device. It is not uploaded anywhere. You never type a diameter or set the scale by hand.
No printer nearby? Try a sample photo
These are synthetic renders (computer-generated, labelled as such in the test set). They run through exactly the same measurement code as your own photo.
What this can and cannot do
Supported conditions
- One ring and the printed 20 mm marker on the same flat, white sheet, both fully visible.
- Near-overhead photo: camera within about 10° of straight down. Steeper photos are rejected.
- Sharp focus: edge blur on the marker under 0.65 mm, and at least 5 pixels per millimetre.
- The ring's opening must show paper, with a visible inner edge around at least half of the circle.
- Plain round bands. Stones, settings or bent rings that stop the ring lying flat are not supported.
How the measurement works
- Scale. The marker's four edges are located to a fraction of a pixel. Its known 20 mm side gives millimetres per pixel and the tilt of the paper.
- Flatten. The photo is mathematically flattened to a straight-down view of the paper using the marker.
- Inner edge. 240 rays from the centre of the opening find the paper-to-ring edge. The largest set of edge points that agree on one circle is the inner boundary.
- Checks. Missing marker, blur, steep angle, low resolution or an unclear edge stop the measurement and ask for a retake.
Measurement uncertainty
The ± value is a model, not a guarantee. It combines pixel resolution, edge blur, scatter of the inner edge, marker localisation, residual perspective, the distance between marker and ring, and the ring's height.
The ring's height matters most: the upper rim sits closer to the lens than the paper, so it looks slightly larger and can hide part of the opening. The model assumes a 4 mm tall band and never reports better than ±0.2 mm. Wide bands (over about 5 mm) photographed from close up can read high by more than the stated value.
Accuracy has only been checked on the test set listed in the repository. Treat the result as an estimate to within about one ring size, not a jeweller's measurement.
Sizing table, source and rounding rule
Table: ISO 8653:2016, Jewellery — Ring-sizes — Definition, measurement and designation. The ISO ring size is the inner circumference in millimetres.
Rounding rule: size = π × inner diameter, rounded to the nearest whole millimetre (exact halves round up). When the uncertainty spans more than one size, the range is shown.
US/Canada equivalent (secondary): inner diameter = 11.63 mm + 0.8128 mm × size, rounded to the nearest half size. Source: published North American scale as given in Wikipedia, "Ring size".
ISO source: iso.org/standard/62783.html. Table covers ISO 41–76; outside it only the diameter is reported.
| ISO size | Inner Ø (mm) | Ø range rounded to this size | US ≈ |
|---|
Time and cost per image
All computation runs in your browser, so there is no per-image variable cost: no recognition API, no model inference fee, no paid intermediary, and retakes cost nothing. The processing time shown with each result is measured on your device.
Hosting is separate: static files on Cloudflare. The one-off download of the measurement engine is about 13 MB uncompressed and is cached by the browser afterwards.