Common Object Detection

Common object detection adds a traditional object detection model to the processing pipeline. This provide benefits in some situations where people, vehicles or bags are being detected. In those scenarios the common object detection can locate possible objects and the EyesOnIt Large Vision Model can verify that the objects match the provided object descriptions. Common object detection provides an optional pre-processing step before evaluation of object descriptions. This step can help focus the Large Vision Model on the areas of greatest interest.

It is important to understand the limitations of a traditional object detection model as compared to the EyesOnIt Large Vision Model. Traditional object detection models perform very well in many situations, but sometimes struggle when the object to detect is partially obscured or disguised. For that reason, common object detection is not appropriate for all scenarios even when the object type is supported by the object detection model. Some trial and error may be needed to determine whether common object detection is right for your use case.

The settings for common object detection are shown below:

To enable common object detection, select the "Detect Common Objects" checkbox. Then, select the object type that you want to detect. You can select multiple object types. For the desired object types, enter a threshold from 10 to 100. If the object detection model detects the select object type with a confidence level above the selected threshold, the EyesOnIt Large Vision Model will evaluate that detection to see if it matches the provided object descriptions. We recommend setting the thresholds low to ensure that anything that looks remotely like the object type gets evaluated by the Large Vision Model. The default thresholds of 30 are a good starting point. In the example above, vehicles detected with greater than 30% confidence will be evaluated by the Large Vision Model.

The image below shows an example of a vehicle detected by the common object model. One benefit of using common object detection is the ability for EyesOnIt to put a bounding box around the detected object when providing alerts.