SONOSCENE360
收藏资源简介:
SONOSCENE360是由马里兰大学帕克分校团队构建的创新型音频-视觉场景数据集,包含68条经过严格校准的360°全景视频与Ambisonic空间音频配对数据,覆盖喷泉、厨房等六类真实场景。该数据集通过专业设备(Insta360 X5相机和RØDE NT-SF1麦克风)采集,标注了声源的语义标签、方向描述及空间坐标,为3D音频-视觉联合生成任务提供基准评估。其核心价值在于首次实现了视觉场景几何结构与空间声场的精准对齐,支持自由视点探索与跨模态语义一致性研究,可应用于VR沉浸式体验、单样本声学学习等前沿领域。
SONOSCENE360 is an innovative audio-visual scene dataset constructed by the team from the University of Maryland, College Park. It contains 68 pairs of rigorously calibrated 360° panoramic videos and Ambisonic spatial audio data, covering six types of real-world scenes including fountains and kitchens. Collected via professional equipment: the Insta360 X5 camera and the RØDE NT-SF1 microphone, the dataset is annotated with semantic labels, directional descriptions and spatial coordinates of sound sources, serving as a benchmark for evaluating 3D audio-visual joint generation tasks. Its core value lies in the first precise alignment between the geometric structure of visual scenes and spatial sound fields, which supports free viewpoint exploration and cross-modal semantic consistency research. It can be applied to cutting-edge fields such as VR immersive experiences and one-shot acoustic learning.




