Key takeaways
- A size recommendation built on a bad body estimate is worse than none, it carries false confidence, so accuracy is the whole point.
- The field spans five approaches, from camera-free questionnaires to depth-sensor capture, each trading friction against precision.
- Capture conditions, lighting, pose, clothing, scale reference, move the error bar as much as the algorithm does.
- Honest benchmarking reports ranges under stated conditions; a single hero accuracy figure is a sales pitch, not a measurement.
- Accuracy is only half the job: a precise body estimate still needs precise garment data on the other side of the match.
Accuracy is the whole point
A body scan is only useful if the measurements are right. A size recommendation built on a bad estimate is worse than no recommendation, because it carries false confidence, the shopper trusts it, orders the wrong size, and now distrusts the whole category of tools, not just yours.
So the question is not whether mobile scanning is impressive. It is how accurate each approach is, under what conditions it holds up, and what happens to the recommendation when conditions are imperfect, because on a live product page, conditions are always imperfect.
The technologies compared
- Photo-based scanning: two phone photos analysed by computer vision. Widely accessible, no special hardware.
- Depth-sensor scanning: uses LiDAR or structured light on capable phones for richer geometry.
- Single-image estimation: one photo only, lower friction but more inference required.
- Wearable or app-guided measurement: shopper-assisted capture for tighter inputs.
- Questionnaire-based profiling: no camera at all, fast, dependent on honest answers.
The trade-off that governs them all
Every approach sits on the same curve: precision against friction. Depth sensors capture the richest geometry but exclude shoppers without capable hardware. Single-image estimation asks the least and infers the most, which widens the error bar. Questionnaires ask nothing of the camera but everything of the shopper's self-knowledge, and self-reported measurements are notoriously off.
Two-photo scanning is the pragmatic middle: enough signal to resolve real proportions (the side view supplies the depth a front view must guess), on hardware every shopper already owns. That is why it has become the workhorse of production sizing, the approach behind flows now running at the scale of millions per year.
What moves the error bar
Accuracy depends on more than the algorithm. Lighting, pose, clothing fit during capture, and a reliable scale reference all swing the result. Controlled capture beats casual capture every time, which is why good products engineer the capture flow itself: clear pose guidance, automatic quality checks, a retake prompt when the frame will not support a solid estimate.
Honest benchmarking reports a range, not a single hero figure, and states its capture conditions. Any vendor quoting one perfect accuracy number across all conditions is selling, not measuring. The right questions to ask: accuracy on which measurements, at which percentiles of the population, under whose capture conditions, validated against what ground truth?
Accuracy is only half the job
A precise measurement still needs something to compare against. The recommendation is only as good as the garment data on the other side: the item's real spec and how its fabric behaves. A perfect body estimate matched to a vague brand chart reproduces the chart's failure, as we argue in Size Charts Are Lying to Your Customers, the chart is usually the weakest link.
That is why Smart Sizing pairs the scan with garment-specific measurements, and why the commercial results, an 11% return cut at Showroomprivé, −16% size-related returns in ETAM's A/B test, come from the match, not from either half alone. A strong body estimate matched to a strong garment spec is what produces a size the shopper can trust.
How accurate is mobile body scanning?
It depends on the approach and the capture conditions: two-photo computer vision resolves real proportions well on ordinary phones, depth sensors add geometric precision on capable hardware, and single-image methods trade accuracy for lower friction. Credible vendors quote ranges under stated conditions, not one universal figure.
Do I need LiDAR on my phone for a body scan?
No. Two-photo scanning works on any phone camera and resolves depth from the side view; LiDAR-class sensors refine geometry but restrict the audience to specific devices, which matters when the tool must serve every shopper on a product page.
What affects body scan accuracy the most?
Capture conditions: lighting, pose, how fitted the clothing is during capture, and a reliable scale reference. Well-designed flows guide the pose and reject frames that will not support a solid estimate, which narrows the error range more than most algorithmic gains.
Is a body scan enough to recommend a clothing size?
No, the body estimate must be matched against the specific garment's real measurements and fabric behaviour. Production results like ETAM's 16% cut in size-related returns come from that body-to-garment match, not from measurement alone.
Conclusion
Mobile body scanning spans a spectrum, from camera-free questionnaires to depth-sensor capture, each with its own accuracy profile and trade-offs. Judge them on honest, condition-aware benchmarks, and remember the measurement is only half the equation. The garment match is the other half.
