It is normal for the algorithm to be unable to read all the race numbers in all of your photos. Several factors can explain why some numbers are not detected — some can be corrected, while others are due to the inherent uncertainties of any artificial intelligence solution.
Poorly defined excluded words #
If you have entered values in the “Words to exclude” field in your album settings, check that none of the values are mistakenly excluding valid bib numbers. An overly broad exclusion may filter out legitimate numbers in addition to the false detections you are targeting.
💡 Also check the min/max range of the bib number — if it is too restrictive, valid numbers outside the range will be ignored even if they have been read correctly.
The face and the bib number are not aligned in the same photo. #
The Face + Bib algorithm only associates a face with a bib when they appear simultaneously in the same photo. If the face is visible but the bib is out of frame, folded, covered or blurred — or vice versa — the association cannot be made.
- Ensure that your photographs clearly capture the face and race number in the same frame.
- For photos where only the race number is visible, identification will be partial — the participant will be recognised but without an associated facial image.
- Organise a photocall to enrich facial prints regardless of the legibility of the bib — see the article "Using photocalls".
Shooting conditions #
Certain situations impair the legibility of the bib and may prevent it from being detected:
- Race number creased, folded or partially obscured by clothing, an arm or equipment
- Blurred race number — shutter speed insufficient to freeze motion
- Race number in backlight or in a marked shadow area
- Race number too small in the frame — shot too wide, making the number illegible
- Back or side angle — the bib is not facing the camera
The vagaries of artificial intelligence #
Even under optimal conditions, the algorithm may not recognise certain bib numbers. AI is not infallible — unusual typography, difficult contrasts, decorative fonts, or damaged bibs can cause the algorithm to fail even on good-quality photos.
These residual cases are normal and expected. They can be corrected manually — see the article “Manually adding a bib number”.
What should be done about unrecognised race numbers? #
- Use the 'Unidentified' filter in 'My Photos' to isolate photos without identification.
- Correct manually by entering the bib number from the photo file — see the article "Manually adding a bib number".
- Use keyboard shortcuts to speed up serial input — see the article "Keyboard shortcuts and fast input".
💡 A recognition rate of 80 to 90 per cent is a good result for a standard sporting event. The remaining 10 to 20 per cent can be corrected manually or left as is if the volume justifies it — participants can always search for their photos manually in the gallery.
