Projects · Minkuk Kim

Streaming video understanding

TimeTraveler: a question can point to past, current or future evidence in a video stream; existing memory-based systems only look back.

TimeTraveler NeurIPS 2026 Look back, look now, or wait: planning where the evidence for a question lives in a live stream. Project page

Long-form video question answering

ReQuest concept: uniform and similarity-based sampling miss the conclusive frame; re-thinking with question-understandable frame selection finds it.

ReQuest ECCV 2026 Answer, doubt, look again: going back only for the frames a question actually needs. Project page

Memory-augmented dense video captioning

HiCM² motivation: human recall from memory versus model retrieval from a hierarchical memory.

CM² · HiCM² CVPR 2024 · AAAI 2025 Recalling past descriptions, then organizing that memory in layers, to find and caption events. Project page