Human3.6M-C, HumanEva-I-C

本研究开发了两个基准数据集Human3.6M-C和HumanEva-I-C，用于评估视频基3D姿态提升器的鲁棒性。这些数据集通过在现有视频中加入如临时遮挡、运动模糊和像素级噪声等常见视频损坏，模拟真实世界视频中的挑战。数据集的创建旨在解决现有3D姿态估计技术在面对视频损坏时的性能下降问题，特别是在动作识别、虚拟/增强现实、人机交互和医疗健康等领域的应用。

This study developed two benchmark datasets, Human3.6M-C and HumanEva-I-C, for evaluating the robustness of video-based 3D pose enhancers. These datasets simulate the challenges in real-world videos by adding common video corruptions such as temporary occlusions, motion blur, and pixel-level noise to existing video data. The development of these datasets aims to address the performance degradation of existing 3D pose estimation techniques when subjected to video corruptions, especially for their applications in fields including action recognition, virtual/augmented reality, human-computer interaction, and healthcare.