Data from a real wedding: QR codes and face recognition
Data·Last updated: ·By Ali Mert Güleç
Short answer
At a wedding held with Photonza on 26 September 2026, 75 guests joined by QR code and 48 of them (64%) searched for their own photos with a selfie. The photographer's 220 shots contained 1,603 faces, a median of 6 per photo. After the photos were uploaded, 47 of the 48 guests found their own photos (median 10 each). Most guests (59 of 75) joined in the days after the wedding, not on the night.
Summary
| Measure | Value |
|---|---|
| Guests who joined by QR code | 75 |
| Searched by selfie (Find Me) | 48 (64%) |
| Photographer shots | 220 |
| Faces detected | 1,603 |
| Faces per photo (median / max) | 6 / 27 |
| Face width (median, share of photo width) | 4.4% |
| Photos matched to at least one guest | 179 / 220 |
| Find Me users who found their own photos | 47 / 48 |
| Photos found per person (median) | 10 |
| Wrong person among 10 random guests checked by hand | 0 |
When did guests join?
| Day | Guests joined |
|---|---|
| 26 Sep (wedding) | 16 |
| 27 Sep | 28 |
| 28 Sep (photos uploaded) | 23 |
| 29 Sep | 6 |
| 30 Sep | 2 |
Takeaway: four in five guests joined after the wedding. Re-sharing the QR code in a thank-you message and when photos are uploaded is the step that brings the most guests in.
How many faces per photo?
| Faces | Photos |
|---|---|
| 1 | 2 |
| 2–3 | 36 |
| 4–6 | 84 |
| 7–10 | 54 |
| 11–20 | 39 |
| 21 or more | 5 |
Finding yourself by eye across 220 photos means looking at 1,603 faces. Face recognition does it in under a second per search (0.9 s for a 500-result search in our scale test).
How many selfies?
29 of the 48 guests added one selfie and 19 added more than one (up to 5). More selfies help with photos taken from different angles.
Method
- Faces were detected with the open-source YuNet model and turned into 128-number signatures with SFace; similarity is cosine similarity.
- Thresholds were tuned on this wedding: "sure" 0.48 (with face-quality checks; 0.55 and above without), "possible" 0.42. Each face goes to the guest it resembles most.
- Manual check: every returned photo of 10 random guests was inspected; none showed the wrong person.
- Privacy: the tables hold totals only. No names, face images or per-person rows are published. Photos and face data are deleted 30 days after the event ends.
- Limits: one wedding, one photographer. Results may differ with other lighting, crowds and shooting styles.
Download the raw table: real-wedding-2026.csv (CC BY 4.0; free to use with attribution).
Frequently asked questions
Which wedding is this data from?
A wedding held with Photonza in Türkiye on 26 September 2026. Names are not published to protect the couple's and guests' privacy.
Can I use the data?
Yes. The CSV is licensed CC BY 4.0; use it with the credit "Photonza, photonza.com".