Lightweight unsupervised learning framework of visual representation for mobile applications

Theme

Image and Video

Topic

Lightweight unsupervised learning framework of visual representation for mobile applications

Background

With the development of 5G, Internet information content is changing rapidly, such as video, 3D data. How to make full use of these new Internet content to better understand products is an important technical direction for Alibaba; on the other hand, the emergence of the new Internet content also makes the cost of data understanding and annotation more expensive. 

This project attempts to explore the general visual representation by using self -supervised learning method, which can make the learned semantic features have better generalization representation, and can quickly transfer to a variety of subsequent specific tasks, for example, object recognition or detection, and reduce the cost of subsequent model development.

The self -supervised learning method will be used in mobile applications, for example photo filter, photo denoising, photo inpainting and super resolution generation.  

Target

  • Developing photo fllter, photo denoising, photo inpainting and super resolution generation on mobile end.
  • ≥1 paper published in the CCFA recommended catalogue. 

Related Research Topics

  • Self -supervised learning
  • Mobile CNN
  • Image Generation and Inpainting

 

Suggested Collaboration Method

AIR (Alibaba Innovative Research), one-year collaboration project. 

 

Scan QR code
关注Ali TechnologyWechat Account