Tencent AVS: A Holistic Ads Video Dataset for Multi-Modal Scene Segmentation

Temporal video segmentation and classification have been advanced greatly by public benchmarks in recent years. However, such research still mainly focuses on human actions, failing to describe videos a holistic view. In addition, previous tends pay much attention visual information yet ignores the multi-modal nature of videos. To fill this gap, we construct an ‘Ads Video Segmentation’ dataset (AVS) ads domain escalate analysis new level. AVS describes from three independent perspectives as ‘presentation form’, ‘place’, ‘style’, contains rich video, audio, text. is organized hierarchically semantic aspects for comprehensive temporal with levels categories multi-label classification, e.g., ‘place’ - ‘working place’ ‘office’. Therefore, distinguished datasets due its information, view categories, hierarchical granularities. It includes 12,000 videos, 82 classes, 33,900 segments, 121,100 shots, 168,500 labels. Accompanied AVS, also present strong baseline coupled class prediction. Extensive experiments are conducted evaluate our proposed method well existing representative methods reveal key challenges AVS.