HaGRID

HaGRID (手势识别图像数据集) 是一个大型图像数据集。可用于图像分类或图像检测任务，适用于视频会议、智能家居、智能驾驶等场景。
海格的大小为716gb，数据集包含552,992幅FullHD (1920 × 1080) RGB图像，分为18类手势。数据分为92% 训练集和8% 测试集user_id，其中509,323图像用于训练，43,669图像用于测试。
该数据集包含34,730个唯一人物和至少该数量的唯一场景。受试者为18至65岁的人。该数据集主要收集在室内，照明变化很大，包括人造和自然光。数据集还包括在极端条件下拍摄的图像，例如面对和面对窗口。此外，受试者必须在距相机0.5到4米的距离处显示手势。

HaGRID (Hand Gesture Recognition Image Dataset) is a large-scale image dataset suitable for image classification and object detection tasks, applicable to scenarios such as video conferencing, smart home, and autonomous driving. The dataset has a total size of 716 GB, containing 552,992 FullHD (1920 × 1080) RGB images categorized into 18 gesture classes. The data is split into a 92% training subset and an 8% test subset, with 509,323 images allocated for training and 43,669 images for testing. It includes 34,730 unique individuals and a corresponding number of unique scenes. The subjects are aged between 18 and 65 years old. Most of the dataset was collected indoors with significant lighting variations, covering both artificial and natural light. The dataset also contains images captured under extreme conditions, such as facing toward or away from a window. Additionally, subjects were required to perform gestures at a distance of 0.5 to 4 meters from the camera.

搜集汇总

背景与挑战

背景概述

HaGRID是一个大型手势识别图像数据集，包含552,992张FullHD RGB图像，分为18类手势，适用于图像分类和检测任务，如视频会议和智能家居。数据集规模达716GB，训练集占92%，测试集占8%，涉及34,730个唯一人物和多样室内场景，照明条件变化大，包括极端环境。发布机构为SberDevices，于2022年发布，支持相关研究和应用开发。