EPIC-KITCHENS-55

EPIC-KITCHENS-55数据集包括一组432个以自我为中心的视频，由32名参与者在他们的厨房中以60fps的速度与头戴式摄像机记录。对于在厨房中自由进行与烹饪，食物准备或洗碗有关的活动的参与者，没有指导脚本。每个视频被分成具有特定开始和结束时间的短动作段 (平均持续时间为3.7s) 以及描述动作的动词和名词注释 (例如，“打开冰箱”)。动词类是125的，名词类是331的。数据集分为一个火车和两个测试拆分。

EPIC-KITCHENS-55 dataset comprises 432 egocentric videos captured by 32 participants using head-mounted cameras in their home kitchens at 60fps. No predefined scripts were provided to participants, who freely engaged in cooking, food preparation or dishwashing-related activities within the kitchen environment. Each video is segmented into short action clips with specific start and end timestamps, with an average duration of 3.7 seconds, paired with verb and noun annotations describing the corresponding actions (e.g., "open the refrigerator"). There are 125 verb classes and 331 noun classes in total. The dataset is split into one training split and two test splits.

搜集汇总

背景与挑战

背景概述

EPIC-KITCHENS-55是一个行为识别数据集，包含432个以自我为中心的视频，由32名参与者在厨房中自由录制，涉及烹饪等活动。这些视频被分割为短动作段，并标注了动词和名词类别，用于动作识别任务。