How to parse StoryBench training data annotations?

[Wuziyi616]

Hi, I'm trying to process the provided training data annotations. For example, I want to parse uvo_dense-train_pipeline.json into the same format as e.g. data/uvo-test.json, which includes sentence_parts, start_times, end_times, and background_description. Is there a script that can do so?

I'm a bit confused by what each data entry in uvo_dense-train_pipeline.json mean. Is the caption term equivalent to the sentence_part? And keyframe_selection_indices indicate the timestamps of this segment?