Labels and Smart Filters

Understanding Labels in HelperBot Explainability

HelperBot’s new Explainability features provide a smarter, more transparent way to evaluate ID photos. At the heart of this improvement is what we call Labels—a powerful and flexible system that allows HelperBot to provide precise, actionable feedback.

What Are Labels?

When a photo is submitted, HelperBot doesn’t just look at it as a whole. Instead, it runs multiple classifiers—each one trained to detect a specific photo attribute, such as lighting quality, glare, background clutter, or whether the face is clearly visible.

Each classifier returns a confidence score about whether a particular condition or issue is present. These conditions are then marked as Labels. Think of each label as a data point that helps paint a fuller picture of what's going on in the photo.

What Do We Do With Those Labels?

This is where the magic happens. Once the labels are in place, HelperBot checks Photo Requirements to see if any Smart Filters need to be activated to filter out the submitted photo. These Smart Filters are customizable rules that determine how the system should respond, depending on the label's confidence score.

Each filter can be configured to respond in one of several ways:

• Automatically approve the photo
• Automatically deny the photo
• Recommend approval
• Recommend denial
• Require human review
• Make no suggestion

How Smart Filters Work

Each label can trigger a Smart Filter based on the confidence score thresholds—a high threshold, a low threshold, or a range between the two. Based on the score, the Smart Filter will activate and carry out its assigned action.

For example:

• If the "Face Visibility" label has a low score, it may trigger a denial, effectively filtering out the bad photo.
• If the "Approvable ID Photo" label has a score within an acceptable range, it may recommend approval.
• If a label score is ambiguous, the system may simply require human review or make no recommendation.

Multiple Smart Filters can be defined for each photo requirement, offering fine-grained control over how HelperBot handles various edge cases and quality concerns.

Why This Matters

This label-based infrastructure gives HelperBot a much deeper understanding of the submitted photo and gives you, the user, more visibility into why a photo was accepted or denied. It also provides card offices with the tools they need to fine-tune HelperBot’s behavior to match specific institutional requirements or tolerance levels.

Video Explanation

Questions? We're Here to Help

If you're curious about configuring Smart Filters for your organization or want to learn more about the classifiers and labels in use, contact our support team at support@cloudcard.us.