I am testing recognition with masks and finding false positive matches that is raising concerns for their use case. How can I reduce false recognitions.
With masked faces, there is very little information for SAFR to differentiate two false with. Therefore, SAFR may show a higher percentage of false positive matches. You can increase the recognition threshold specific to mask faces as follows to reduce the chance of a false positive.
To reduce number of false positive with masked faces you can make the following adjustments:
- Reduce “Masked face threshold offset” slightly (e.g. to -0.05 or -0.1)
- Increase “Minimum recognitions to lock onto identity” to 2 or 3
Following shows these settings in SAFR Desktop Preferences.