Question regarding Object Interaction task of STAR & Charade dataset

[cmh1027]

    {
        "video": "K9UXS.mp4",
        "question": "Which object was tidied up by the person?",
        "candidates": [
            "The table.",
            "The clothes.",
            "The towel.",
            "The blanket."
        ],
        "answer": "The blanket.",
        "start": 2.65134176537509,
        "end": 30.34865823462491,
        "accurate_start": 22,
        "accurate_end": 31
    },

I checked that "accurate_start" and "accurate_end" fields came from "start", "end" field of original STAR annotation. Then, where did 'start', 'end' file come from? I can't find any information from the original annotation.

• Same situation in action localization of Charade dataset

    {
        "video": "ONMCW.mp4",
        "question": "In the given video, when does the action 'person takes a glass from the desk' take place?",
        "candidates": [
            "At the end of the video.",
            "At the beginning of the video.",
            "In the middle of the video.",
            "Throughout the entire video."
        ],
        "answer": "In the middle of the video.",
        "start": 15.299999999999997,
        "end": 33.6,
        "accurate_start": 21.4,
        "accurate_end": 27.5
    },

[Andy1621]

Good question! The start and end are randomly generated based on the accurate start and end, since we hope to improve the question difficulty.