Ego4D Episodic Memory Benchmark
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/Yisen-Feng/ObjectNLQ
下载链接
链接失效反馈官方服务：
资源简介：
该数据集为评估模型在长视频序列中对动作定位的性能提供了一个基准，尤其是通过文本查询的方式。数据处理时，以16帧的步长处理视频特征，并且为了集成学习，训练集和验证集被划分为五个折叠。该数据集的任务是自然语言查询及目标步骤定位。
This dataset serves as a benchmark for evaluating model performance on action localization in long video sequences, particularly via text-based querying. During data processing, video features are processed with a stride of 16 frames. For ensemble learning, the training and validation sets are split into five folds. The task of this dataset is natural language querying and target step localization.
提供机构：
Ego4D
