Efficient Semantic-Guidance High-Resolution Video Matting

Video matting has made significant progress in trimap-based field. However, researchers are increasingly interested in auxiliary-free matting because it is more useful in real-world applications. We propose a new efficient semantic-guidance high-resolution video matting network for human body. We apply the convolutional network as the backbone while also employing the transformer in the encoder, which is used to utilize semantic features, while ensuring that the network is not overly bloated. In addition, a channel-wise attention mechanism is introduced in the decoder to improve the representation of semantic feature. In comparison to the current state-of-the-art methods, the method proposed in this paper achieves better results while maintaining the speed and efficiency of prediction. We can complete the real-time auxiliary-free matting for high-resolution video (4K or HD).