MatAnyone achieves superior coherent memory propagation capabilities through Region Adaptive Memory Fusion technology. This technology automatically remembers the features of the previous frame and fuses them into the current frame processing, ensuring that the keying results in a video sequence remain highly coherent in the time dimension. Compared to traditional frame-by-frame processing, this technique effectively solves common problems such as inter-frame flickering.
In practice, users only need to provide the first frame mask, and the program will automatically complete the memory propagation processing for subsequent frames. The research team also provides adjustable parameters such as memory_fusion_rate in the configuration file, which makes it easy for the user to optimize the effect according to the video characteristics (e.g., sudden lighting changes).
This answer comes from the articleMatAnyone: Extract video to specify the target portrait of the open-source tool to generate the target portrait videoThe
