The Sound of Pixels
This work addresses the challenge of grounding sounds in vision for applications like adjusting sound source volumes, representing an incremental advance in multimodal learning.
The paper tackles the problem of learning to locate image regions that produce sounds and separate input sounds into components from each pixel by leveraging unlabeled videos, achieving improved source separation performance on the MUSIC dataset.
We introduce PixelPlayer, a system that, by leveraging large amounts of unlabeled videos, learns to locate image regions which produce sounds and separate the input sounds into a set of components that represents the sound from each pixel. Our approach capitalizes on the natural synchronization of the visual and audio modalities to learn models that jointly parse sounds and images, without requiring additional manual supervision. Experimental results on a newly collected MUSIC dataset show that our proposed Mix-and-Separate framework outperforms several baselines on source separation. Qualitative results suggest our model learns to ground sounds in vision, enabling applications such as independently adjusting the volume of sound sources.