Introduction
VIDIZMO uses FairFace as the AI model for attribute detection in order to identify individuals with certain qualities. It enables you to search for specific people in your media or evidence files using predefined attributes and their categories. Once the AI model detects a face, it will make predictions and classify it accordingly.
The AI model utilized by the VIDIZMO Indexer App enables this functionality across your Portal. You can also use attribute detection to make your redaction activities efficient and accurate.
Concept
Attribute Prediction using FairFace is an enhancement to the Face Detection capabilities offered by VIDIZMO. In addition to face detection, the AI model now performs classification on the faces it detects based on three attributes: age, gender, and race. By running face detection, you can search for and identify specific individuals in your media or evidence via attribute filters. You can view these options whenever you open a video to view its AI insights.
To learn how to enable video Insights, visit How to Enable and View Video Insights.
You can also view and filter detections using attributes in Studio Space. Moreover, you can even perform redaction on these detected faces by creating filter criteria that meet your preferences. Age is the first attribute you can use to filter out individuals from the detections. In this feature, the available age range is 0 – 70+ years, divided into brackets with specific years. The age brackets are 0-3,4-9,10-19 and 20- 29; these continue till 60- 69, and then the large age bracket consists of ages that are 70+. Selecting one or more age brackets will only bring forward the detections that the model predicts belong to the age range.
The second attribute to filter detections is gender. It has two categories: a person can be classified as male, Male, and Female. Selecting either of the options will only highlight or bring forward the detections that the model predicts are men or women.
The third attribute that can be predicted is race. You can identify individuals from seven races: White, Black, Indian, East Asian, Southeast Asian, Middle Eastern, and Latino Hispanic.
Under each of the attributes, you will sometimes find a Not Predicted category. This category indicates that the model detected a face but could not make a prediction on the respective attribute. You can also use this category as a filter to bring forward unidentifiable individuals in your media or evidence file.
Detection Settings
The detections or predictions made by the AI model depend on your detection settings. One such setting is the Confidence threshold, which is set to 45 by default in your Portal's App Settings. For Face Detection, the confidence threshold determines which object detections are identified as 'Faces.' All object detections have their confidence score, which the AI model uses for identification.
Users can increase or decrease the confidence threshold from the default value of 45 to adjust the number of detections. For instance, if the user sets the confidence threshold to 60, any object detection with a confidence score below 60 will not be classified as a 'Face.'
Increasing the confidence score from the App Settings or the Studio Space will decrease the number of detections, which leads to a possible decrease in predictions as the model has fewer faces to predict from. Similarly, reducing the confidence score will increase the number of face detections, which may increase the number of predictions as the model has more faces to predict attributes.
Attribute Prediction also introduces a new Detection Setting called Face Processing Threshold. It determines which Face Detections the model will consider to make attribute predictions. For instance, if a detected face has a threshold of 80, but the default Face Processing Threshold is set to 86, then the AI model will not consider that face for attribute predictions. It will instead classify that face as Not Predicted. In Studio Space, the Face Processing Threshold appears next to the face detections on the left pane.
The default Face Processing Threshold is 85, with a tolerance of 3. The value of the Face Processing Threshold can be set from 70 to 90. Manager+ users can adjust the Face Processing Threshold in App Settings after selecting an Attribute (or multiple) for prediction. Decreasing the Face Processing Threshold from its default value will enable the model to make predictions on blurry or obscured faces and increase the number of overall predictions. However, it is recommended that the default value of 85 is used for the most accurate attribute prediction results.
Use Cases
Law Enforcement
One of this feature's most prominent use cases would be accurately identifying individuals from evidence files. Law enforcement agencies can save time analyzing security, CCTV, or dashcam footage. If they have a description of the suspect, such as their supposed age, gender, or race, they can utilize attribute filters to yield effective results.
Demographic Analysis
Person attribute detection allows for a more insightful demographic analysis when addressing bias or discrimination. Organizers, surveyors, or researchers can analyze and obtain meaningful information about the treatment of people in a particular field or workplace. Demographic analysis with attribute prediction can highlight which groups face discrimination or have a systematic bias against them.
Consumer Analysis
This feature can also help sellers or retailers identify their audience better. By analyzing the activity in their store, sellers can determine which group spends the most time and makes the most purchases of their products. Getting insights such as these can aid them in tuning their marketing strategy or introducing new products that meet the needs of their most active customers.
Up Next
To enable this feature in the VIDIZMO, refer to our guide "How to Generate a Person's Attribute Insights in VIDIZMO" and
"Utilizing Face Attributes to Apply Redaction in Studio Space."