ImageNet-VidVRD
收藏资源简介:
ImageNet-VidVRD 数据集包含从 ILVSRC2016-VID 数据集中根据视频是否包含清晰的视觉关系选择的 1,000 个视频。它分为 800 个训练集和 200 个测试集,涵盖 35 个类别的常见主语/宾语和 132 个类别的谓词。有 10 人对数据集进行了标注,包括对象轨迹标注和关系标注。由于 ILVSRC2016-VID 数据集已经有 30 个类别的对象轨迹注释,我们通过标记其余 5 个类别来补充注释。为了节省关系标记的工作,我们标记了训练集中视频的典型片段和测试集中的整个视频。
The ImageNet-VidVRD dataset consists of 1,000 videos selected from the ILVSRC2016-VID dataset based on whether the videos contain clear visual relationships. It is split into 800 training videos and 200 test videos, covering 35 common subject/object categories and 132 predicate categories. A total of 10 annotators participated in the labeling work, which includes object trajectory annotation and relationship annotation. Since the ILVSRC2016-VID dataset already has object trajectory annotations for 30 categories, we supplemented the annotations by labeling the remaining 5 categories. To reduce the workload of relationship labeling, we only labeled typical segments of videos in the training set and full videos in the test set.




