Ego2Hands

Ego2Hands数据集由杨百翰大学开发，专注于无约束环境下的双手分割与检测任务。该数据集包含约188,362条标注帧，通过半自动标注和颜色不变合成技术，解决了传统数据集在规模和多样性上的限制。数据集内容涵盖了广泛的手部位置、姿态、肤色和光照条件，通过绿色屏幕设置自动获取分割掩码。创建过程中，使用了22名具有多样肤色的参与者进行自由手部动作录制，以确保数据的多样性和真实性。Ego2Hands数据集的应用领域包括人机交互、活动记录、手势/手语识别以及VR/AR等，旨在解决双手交互场景下的识别与分割问题，提高模型在复杂环境中的泛化能力。The Ego2Hands dataset, developed by Brigham Young University, focuses on the task of two-handed segmentation and detection in unconstrained environments. It comprises approximately 188,362 annotated frames, and addresses the scale and diversity limitations of traditional datasets via semi-automatic annotation and color-invariant synthesis technologies. The dataset covers a broad spectrum of hand positions, postures, skin tones and lighting conditions, with segmentation masks automatically obtained through green screen setup. In the course of its creation, 22 participants with diverse skin tones were recruited to record free-form hand movements, thus ensuring the diversity and authenticity of the dataset. Application scenarios of the Ego2Hands dataset span human-computer interaction, activity recording, gesture/sign language recognition, VR/AR and other fields, with the goal of solving the recognition and segmentation issues in two-handed interaction scenarios and improving the generalization capability of models in complex environments.

搜集汇总

背景与挑战

背景概述

Ego2Hands是一个专注于无约束环境下双手分割与检测的数据集，由杨百翰大学开发，包含约18.8万条标注帧，通过半自动标注和颜色不变合成技术提高数据多样性和规模。它覆盖了多种手部位置、姿态、肤色和光照条件，旨在解决双手交互场景的识别问题，适用于人机交互、手势识别和VR/AR等领域，以增强模型在复杂环境中的泛化能力。