Entity Detection automatically identifies and tracks specific people (faces) and brand logos throughout your video assets. Understanding how MediaSilo AI recognizes these entities and the differences between tracking human faces versus static logos will help you to get the most accurate search results and highlights.
How Entity Detection Works
When you enable AI indexing on a video, MediaSilo analyzes the file and detects faces and logos, turning them into what are known as vectors. A vector is a dynamic data set used by MediaSilo AI to analyze and map the characteristics of that face or logo.
If the AI detects a face it doesn't recognize, it will temporarily label it as an unknown entity with a number, IE: unknown 5. When you click on that unknown face and upload an image and add a name, the system is being told to tag that specific vector with the image and name you provided. Once tagged, the system will instantly apply that person's name everywhere that specific vector appears across your entire workspace. Admins can also preemptively upload images to an Entities Library to establish these vectors before videos are uploaded.
Detection Thresholds: People vs. Logos
When the AI identifies an entity, it compares the vector found in the video against the known vectors in your library. The Detection Threshold determines how close of a match is required for the AI to definitively add to that vector. Users can manually adjust this threshold in real-time to refine their results without needing to reprocess the entire video. A lower threshold is more lenient and will return more results but may include false positives. A higher threshold is stricter and will only return results that closely resemble the known entity.
- People (Faces): Human faces are dynamic. A person might wear heavy makeup, change their hairstyle, or be filmed from various angles. If your detection threshold is set too high, the AI may miss a person if they look slightly different. If the threshold is too low, you increase the risk of false positives. For Example: The AI matching two different people because they have a similar nose. Because of these dynamic variables, a medium threshold is usually the best selection for consistent facial recognition.
- Logos: Unlike people, logos are highly static and standardized. Because their design is fixed, a higher detection threshold is generally better for logos. If your threshold is too low, the AI might mistakenly identify a similar shape or color block as a specific brand.
Best Practices for Uploading Entity Images
Whether you are uploading an image to identify a person or a logo, the framing of your reference image is critical to the AI's success. For the highest rate of success, utilize the built-in zoom and crop tool when uploading an image to ensure the face or logo is front and center, and clearly visible. For more information see: MediaSilo AI - Entity Cropping & Reference Images
Comments
0 comments
Please sign in to leave a comment.