IS3+
收藏资源简介:
IS3+是一个视觉声源定位的改进数据集,扩展了IS3数据集,包括了负音频情况下的评估,例如噪音、静音和屏幕外的声音。数据集包含3240张图像,每个图像配对两个音频样本,共计6480个音频视觉实例。图像通过扩散模型合成,模拟多个物体在图像中的不同大小。音频样本来自Adobe Sound Effects和IS3数据集的清洁样本,以确保音频与图像内容相匹配。
IS3+ is an improved dataset for visual sound source localization, which extends the original IS3 dataset by incorporating evaluations under negative audio conditions such as noise, silence, and off-screen sounds. This dataset contains 3,240 images, each paired with two audio samples, totaling 6,480 audio-visual instances. The images are synthesized via diffusion models, simulating varying sizes of multiple objects within the images. The audio samples are sourced from Adobe Sound Effects and the clean audio samples from the IS3 dataset, ensuring that the audio content matches the corresponding visual content.

- 1通过巴塞罗那庞贝乌法布拉大学 · 2025年



