We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Collaborative Learning to Generate Audio-Video Jointly

Abstract: There have been a number of techniques that have demonstrated the generation of multimedia data for one modality at a time using GANs, such as the ability to generate images, videos, and audio. However, so far, the task of multi-modal generation of data, specifically for audio and videos both, has not been sufficiently well-explored. Towards this, we propose a method that demonstrates that we are able to generate naturalistic samples of video and audio data by the joint correlated generation of audio and video modalities. The proposed method uses multiple discriminators to ensure that the audio, video, and the joint output are also indistinguishable from real-world samples. We present a dataset for this task and show that we are able to generate realistic samples. This method is validated using various standard metrics such as Inception Score, Frechet Inception Distance (FID) and through human evaluation.
Comments: ICASSP 2021 (Accepted)
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR); Multimedia (cs.MM); Sound (cs.SD); Audio and Speech Processing (eess.AS); Image and Video Processing (eess.IV)
Cite as: arXiv:2104.02656 [cs.CV]
  (or arXiv:2104.02656v1 [cs.CV] for this version)

Submission history

From: Vinod Kumar Kurmi [view email]
[v1] Thu, 1 Apr 2021 01:00:51 GMT (3840kb,D)

Link back to: arXiv, form interface, contact.