REALIMPACT

REALIMPACT是由斯坦福大学创建的大型数据集，包含150,000条真实物体撞击声的录音。该数据集在控制条件下记录了50种日常物体的撞击声，并附有详细的注释，包括撞击位置、麦克风位置、接触力分布、材料标签和RGBD图像。REALIMPACT旨在用于视听学习和缩小模拟与现实之间的差距，通过评估两个基准任务，包括听众位置分类和视觉声学匹配，展示了其作为测试平台的有用性。数据集的应用领域包括虚拟现实、动画和基于学习的模拟框架的训练，旨在解决提高物理模拟真实性的问题。

REALIMPACT is a large-scale dataset created by Stanford University, housing 150,000 recordings of real-world object impact sounds. This dataset captures the impact acoustics of 50 everyday objects under controlled laboratory conditions, paired with detailed annotations including impact location, microphone placement, contact force distribution, material labels, and RGBD images. REALIMPACT is designed for audio-visual learning and bridging the gap between simulation and real-world scenarios, and its utility as a research testbed is validated through two benchmark tasks: listener position classification and visual-acoustic matching. The dataset finds applications in virtual reality, animation, and the training of learning-based simulation frameworks, aiming to address the challenge of enhancing the realism of physical simulations.