Revisiting Human-in-the-Loop Object Retrieval with Pre-Trained Vision Transformers