ActivityNet Adverbs

我们评估了HowTo100M副词的方法，该方法从HowTo100M中的83个任务中挖掘了副词。由于注释是从教学视频的自动转录叙述中获得的，因此它们是嘈杂的; 44% 带注释的动作副词对在视频剪辑中不可见。该数据集包含5,824个片段，其中包含72个动词和6个副词中的动作副词对注释。此数据集的明显限制是它包含的副词数量很少，因此我们从现有的视频检索数据集中创建了三个新的副词数据集: VATEX副词，msr-vtt副词和ActivityNet副词。这些包含更少的噪音和更多种类的副词。

We evaluated our adverb extraction approach, which mined adverbs from 83 tasks within the HowTo100M dataset. Since the annotations were obtained from automatically transcribed narratives of instructional videos, they are noisy; 44% of the annotated action-adverb pairs are not visible in the corresponding video clips. This dataset consists of 5,824 segments annotated with action-adverb pairs covering 72 distinct verbs and 6 adverbs. A notable limitation of this dataset is the small size of its adverb inventory. Accordingly, we constructed three novel adverb datasets from existing video retrieval datasets: VATEX-Adverb, MSR-VTT-Adverb, and ActivityNet-Adverb. These datasets exhibit lower noise levels and a broader range of adverb categories.

搜集汇总

背景与挑战

背景概述

该数据集评估了HowTo100M副词方法，从83个任务中挖掘副词，其注释基于教学视频的自动转录叙述，存在噪声，44%的动作副词对在视频中不可见。它包含5,824个片段，标注了72个动词和6个副词中的动作副词对，为减少噪声并丰富副词类型，还从现有视频检索数据集中创建了三个新副词数据集。