Has anyone implemented this model for visdrone? If so, could you share the code base and pre-trained model