• ReBotNet, an efficient and fast framework for real-time video enhancement, is designed for practical use-cases like live video calls and video streams.
  • The method employs a dual-branch framework and a recurrent training approach to reduce memory requirements, lower computation, and speed up inference time.

Key terms:

  • Recurrent Bottleneck Mixer Network (ReBotNet): An efficient and fast framework designed for real-time video enhancement in practical use-cases such as live video calls and video streams
  • Dual-branch framework: A framework consisting of two branches, one learning spatio-temporal features and another improving temporal consistency
  • ConvNext-based encoder: An encoder used to tokenize input frames along spatial and temporal dimensions
  • Bottleneck mixer: A processing method for abstract tokens in the framework
  • Recurrent training approach: An approach leveraging the last frame's prediction to efficiently enhance the current frame and improve temporal consistency


Research ReBotNet Video Enhancement Real Time Video Calls Video Streams Dual Branch Framework ConvNext-based encoder Bottleneck Mixer Recurrent Training