ActivityNet-QA
收藏资源简介:
ActivityNet-QA是一个大规模、全人工标注的视频问答数据集,由杭州电子科技大学和浙江大学联合创建。该数据集包含58,000个问答对,基于5,800个来自ActivityNet的复杂网络视频。数据集的创建过程涉及从ActivityNet中抽样视频,并通过众包方式生成问答对。ActivityNet-QA旨在通过问答形式深入理解视频内容,特别适用于长视频的细粒度视觉理解和时空推理,为视频问答技术的发展提供了重要的基准。
ActivityNet-QA is a large-scale, fully manually annotated video question answering (QA) dataset jointly created by Hangzhou Dianzi University and Zhejiang University. This dataset contains 58,000 QA pairs based on 5,800 complex web videos sourced from ActivityNet. The dataset construction process involves sampling videos from ActivityNet and generating QA pairs via crowdsourcing. ActivityNet-QA is designed to enable in-depth comprehension of video content via question answering, with particular suitability for fine-grained visual understanding and spatio-temporal reasoning of long-form videos, thereby serving as a pivotal benchmark for the advancement of video QA technologies.

- 1ActivityNet-QA: A Dataset for Understanding Complex Web Videos via Question Answering复杂系统建模与仿真重点实验室 · 2019年



