Yoga Pose Dataset

The in-house Yoga pose dataset consists of 27 individuals, comprising nineteen females and eight males, performing each of the ten Yoga poses, namely Malasana, Ananda Balasana, Janu Sirsasana, Anjaneyasana, Tadasana, Kumbhakasana, Hasta Uttanasana, Paschimottanasana, Uttanasana, and Dandasana. The videos of the Yoga poses are collected in both 1080p and 4K resolution at a rate of 30 frames per second using MI Max and One Plus 5T mobile phones. We captured the videos at various locations like gardens, rooms, certified Yoga centers, and terraces to increase the generality of the dataset and make the dataset more realistic so that the model trained using the dataset would work in a complex real-world environment. It is worth mentioning that our in-house created dataset does not contain any video sample created in the controlled laboratory-like environment with proper illumination. We did this deliberately to enhance the generalization ability of the models trained on our dataset. The individuals have voluntarily participated in data collection and have performed the ten Yoga poses with possible variations. <br>

本自研瑜伽体式数据集（in-house Yoga pose dataset）共包含27名受试者，其中女性19名、男性8名，每名受试者均完成全部10种瑜伽体式，分别为马拉萨纳式（Malasana）、阿南达巴拉萨纳式（Ananda Balasana）、贾努西里萨萨纳式（Janu Sirsasana）、阿南贾内亚萨纳式（Anjaneyasana）、塔达萨纳式（Tadasana）、库姆巴卡萨纳式（Kumbhakasana）、哈斯塔乌塔纳萨纳式（Hasta Uttanasana）、帕西莫托塔纳萨纳式（Paschimottanasana）、乌塔纳萨纳式（Uttanasana）以及丹达萨纳式（Dandasana）。本数据集的瑜伽体式视频均采用小米Max（MI Max）与一加5T（One Plus 5T）手机采集，分辨率涵盖1080p与4K，帧率为30帧每秒。我们在花园、室内房间、认证瑜伽场馆以及露台等多样场景下完成视频采集，以此提升数据集的泛用性并增强其真实感，使得基于该数据集训练的模型可适配复杂真实环境。值得说明的是，本自研数据集未包含任何在可控类实验室环境（具备标准照明条件）下采集的视频样本，我们刻意如此设计以强化基于该数据集训练的模型的泛化能力。所有受试者均自愿参与数据采集，并在完成10组瑜伽体式时保留了动作的自然变化细节。