What are the different brand detection models in junbi?

Brand detection is the automated computer vision technology in junbi that scans your video ads to identify and track every frame where your brand elements appear. By recognizing these assets, the platform can calculate your second-by-second Brand Attention score, which measures whether viewers are actually noticing your brand.

The Strategic Value

While human viewers instantly connect a stylized visual symbol and a written wordmark to the same company, computer algorithms process text and shapes differently. By utilizing advanced brand detection models, junbi maps the exact boundaries of your visual assets. This ensures your final predictive data is highly accurate and directly represents how human eyes process your creative.

Understanding the Three Types of Detection Models

Depending on your brand guidelines and specific account setup, junbi offers three distinct ways to track your brand assets.

1. Text-Based Detection

This model is designed to detect standard, typed text elements. When a text-based model runs, the platform draws a standard rectangular bounding box around your written brand name. This method is incredibly fast and highly accurate for wordmarks.

2. Iconic Detection

This model detects a combination of iconography and text, or visual brand icons on their own. Instead of drawing a simple rectangle, iconic detection uses advanced computer vision called segmentation to map the exact, pixel-for-pixel contour of your visual logo. The platform marks these detections on your dashboard with a precise green shape surrounding the asset.

3. Custom Detection

This is a fully tailored model trained specifically on your unique brand inputs and business assets. Custom models use the same high-resolution segmentation technology as iconic models, but they are not restricted to logos. A custom model can be built to detect and track anything your team needs, such as specific product packaging, brand characters, unique design elements, or key assets.

💡 Pro Tip:

Understand the underlying technology. While standard text detection relies on recognizing letters, both iconic and custom models use pixel-for-pixel segmentation. This means they map the literal shape of your asset, providing unmatched visual tracking accuracy for non-text design elements.