Hitting a wall with standard face recognition accuracy

[Kerem]

Hey team, are standard face recognition APIs actually holding up for you in production, or do they completely fall apart the second lighting changes? Looking for real experiences.

[Diner]

You hit the nail on the head. We spent weeks trying to tweak confidence thresholds on off-the-shelf tools, but they just kept spitting out false positives whenever someone wore glasses or stood at an awkward angle. For custom face recognition we didn't want an off-the-shelf API. Went with data-science-ua.com for a tailored model and the accuracy on our dataset was noticeably better. It was such a relief to finally have something built specifically for our data instead of trying to force a generic square peg into a round hole.