Intelligent Production of 4K HDR Video

Theme

Image and Video

Topic

Intelligent Production of 4K HDR Video 

Background

With the advent of the 5G era, the development of cloud computing and the advancement of display terminals, the ultra-high definition video industry featuring at high resolution, high frame rate, high dynamic range / wide color gamut, has begun to flourish. It has brought the ultimate visual enjoyment to consumers, and will surely usher in an explosion in the next few years. However, currently there exists very few ultra-high definition videos which meet the 4K HDR standard. And the vast majority of videos we see are old, high/standard-definition and lower quality videos. Their common characteristics are low resolution, low frame rate, low dynamic range and narrow color gamut. How to convert these existing videos into high quality 4K HDR format is an urgent task to be solved. In addition, for the production of film and television dramas, the shot materials are also in need of high resolution, high dynamic range and vivid color. They often need to be converted into 4K HDR.

In recent years, with the rapid development of deep learning technology and the tremendous progress of hardware such as GPUs, AI-based technology has flourished in the video processing field. We hope to make use of AI technology in the production of video content with 4K HDR format, to satisfy the urgent market needs, and produce artistic-sense high-quality videos.

Target

  • An algorithm prototype with detailed description documents, which should be capable to solve the critical technical problems.
  • An algorithm implementation code, which should produce correct results without memory leaks and crashes, and be easily deployed on Alibaba Cloud servers.
  • A test report, which
    • should include single and comprehensive tests,
    • must clarify the test objectives, test video sequences and corresponding picture quality results,
    • and must include subjective and objective evaluation metrics.
  • At least one domestic/international patent
  • At least one top conference/journal paper, such as CVPR, ICCV, ECCV, TIP, PAMI, CSVT, IJCV, etc.
  • An outstanding ranking in at least one influential video processing competition.

Related Research Topics

  • To produce high-quality super-resolution video
    • Solve the conversion from high-definition, standard-definition, and lower-resolution video to 4K video, and basically need to support 2 times, 3 times, 4 times, 6 times, and 8 times super-resolution.
    • Reconstruct fine-grained and no-overshooting edges.
    • Recover rich and natural details, especially the fine details of the face.
    • Remove kinds of noises, such as scratches, camera noise, compression noise, spatial-temporal random noise, etc.
  • To convert SDR video to high-quality HDR format
    • Support the conversion from SDR to HDR10, HLG, HDR10+.
    • Support the conversion to multiple peak luminance.
    • Adaptively adjust luminance based on video content, avoid overexposure in highlight areas and keep details in dark areas while suppressing the noise
    • Retouch rich and natural colors and appropriate skin tone
  • Keep the produced video stable and no flickering in spatial and temporal domain

 

Suggested Collaboration Method

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

 

Scan QR code
关注Ali TechnologyWechat Account