VGG-SS, short for VGG Sound Source, is a video dataset for evaluating sound source localization. The dataset contains more than 200 categories, 5,000 videos, and new annotations for the VGG-Sound dataset, which is 10 times larger than existing datasets. The visible sound sources in each video clip are clearly marked with bounding boxes. Unlike Flickr SoundNet, the sound source localization of this dataset is based on videos.