상세정보
학술대회
ActionNet-VE Dataset: A Dataset for Describing Visual Events by Extending VIRAT Ground 2.0
- 발행일
- 201511
- 출처
- International Conference on Signal Processing, Image Processing and Pattern Recognition (SIP) 2015, pp.1-4
- 협약과제
- 15MS4500, (1세부) 실시간 대규모 영상 데이터 이해·예측을 위한 고성능 비주얼 디스커버리 플랫폼 개발, 박경
- 초록
- This paper introduces a dataset for recognizing and describing interactive events between objects of interest including persons, cars, bikes, and carried objects. Although there have been many video datasets for human activity recognition, most of them focus on persons and their actions and sometimes ignore the specific information on related objects, such as their object type and minimum bounding boxes, in annotations. ActionNet-VE dataset was designed to include full annotations on all objects and events of interest occurred in a video clip for describing the semantics of the event. The dataset adopt 75 video clips from VIRAT Ground 2.0, and extend annotations on the events and their related objects. In addition, the dataset describes semantics of each events by using elements of sentences, such as verb, subject, and objects.
- KSP 제안 키워드
- Bounding Box, Human activity recognition(HAR), Video clips, interactive events, visual events