THU-READ
收藏资源简介:
THU-READ数据集是由清华大学开发,专注于识别与手部动作相关的日常行为。该数据集包含40种不同动作,由8名参与者重复3次,总计960个RGB-D视频。本研究从中选取了2124张代表性图像进行像素级标注,用于支持实时语义分割网络的研究,特别是在混合现实应用中,以增强虚拟现实体验的真实感。数据集的创建涉及使用AMT平台进行手动标注,确保了标注的准确性和质量。该数据集的应用领域主要集中在提升混合现实中用户与真实物体交互的自然性和效率。
THU-READ Dataset is developed by Tsinghua University, focusing on recognizing daily behaviors related to hand movements. This dataset contains 40 distinct actions, performed 3 times by 8 participants, totaling 960 RGB-D videos. In this study, 2124 representative images were selected from it for pixel-level annotation to support research on real-time semantic segmentation networks, particularly in mixed reality applications to enhance the realism of virtual reality experiences. The dataset's creation involved manual annotation using the AMT platform, which ensured the accuracy and quality of the annotations. The primary application areas of this dataset center on improving the naturality and efficiency of user interaction with real objects in mixed reality.

- 1Real Time Egocentric Object Segmentation: THU-READ Labeling and Benchmarking Results诺基亚贝尔实验室 · 2021年



