Number of video and clips: how to trim video into clips

[yjhong89]

In AdsQA paper, there is written

Based on these domain-specific advantages of ad videos,
we introduce AdsQA, a comprehensive and carefully curated VideoQA benchmark derived from 1,544 ad videos
containing 10,962 clips, spanning a total of 22.7 hours of
video content. 

I am downloading this datasets but how 10962 clips can get from raw videos?
Is there any trimming information or related code?

Thanks

[xinwei666]

Thank you for your interest. During our data construction process, we used PySceneDetect to segment advertisement videos into multiple clips in order to obtain finer-grained features and synchronized ASR text. Therefore, you may use this tool to split the videos as well. However, under the AdsQA setup, each QA pair does not correspond to a specific clip. You can directly use the entire video as input—in fact, the models we evaluated were implemented in this way.

Additionally, our AdsQA dataset was featured as Challenge 3 in the ICCV MARS2 Workshop and attracted many participants. The winning team, for example, used PySceneDetect to divide the videos into several clips, extracted features from each clip, and then generated answers. You may refer to such methods (https://arxiv.org/abs/2509.14142).