VIS-MODE: Interactive Complex Event Detection from VLM-Based Video Observations