Am using the model person-detection-retail-0013, have tried some others including person-vehicle-bike-detection-2001. What we are seeing is that the model recognizes gaming chairs, posters of people and mannequins as people.
This means that it adds a lot of "garbage" to the tracks or to reidentification which is not required.
These objects are also very much stagnant throughout the day.
Is there any way to exclude these object? Or alternatively to train them out of the model as not a person?
Just to add that every site is different (so not always the same object), but there are stagnant objects at each site that are not "a person".
The only thing we can think of is somehow seeing that the detection area doesn't change over time and ignoring it, however since every frame is a frame on itself there is no real history unless the tracks are explored.
You may use the OpenVINO Training Extension to retrain the pre-trained model: https://github.com/openvinotoolkit/training_extensions
Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question