The neural network model of classifier behavior closeup openvino is trained on a dataset that is clear and visible near the face at a short distance. For other scenarios, you need to collect data specific to the scene and train it before using it.
There could also be other factors that are affecting the model’s performance, such as lighting conditions, camera angle, or other environmental factors. It may be worth exploring other pre-trained models or training your own custom model if you need to detect smoking or drinking behavior for your specific use case.