EPIC-KITCHENS-100

该数据集是一个大规模的第一人称视角数据集，用于预测个人的未来离散动作和连续运动。该数据集包含了122个动作类别（39个动词和83个名词），并且为了确保每个动作至少出现50次，需要对数据进行筛选。数据规模涵盖了4455个示例，来自135个视频。该数据集的任务是进行轨迹和动作预测。

This dataset is a large-scale first-person perspective dataset dedicated to predicting an individual’s future discrete actions and continuous movements. It consists of 122 action categories, including 39 verbs and 83 nouns. Data screening is performed to ensure that each action appears at least 50 times. The dataset contains 4,455 samples sourced from 135 videos, with its targeted task being trajectory and action prediction.

搜集汇总

背景与挑战

背景概述

EPIC-KITCHENS-100是目前最大的第一人称视角（egocentric）视觉数据集，包含100小时的非脚本化录制内容，覆盖45个厨房环境，具有丰富的多语言叙述和详细的注释。数据集特点包括20M帧、90K动作片段、97个动词类和300个名词类，适用于多种计算机视觉挑战任务。