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