Dual Attention Mechanisms-Embedded Deep Learning Method for Video Subtle Motion Magnification

Video subtle motion magnification technology is widely used in engineering monitoring, but deep learning methods often suffer from quality degradation at high magnifications. This paper proposes a dual-attention mechanism (coordinate attention + self-attention) for video motion magnification. It enhances texture representation and reduces artifacts. Experiments show a 3.5dB PSNR and 0.026 SSIM improvement on the Baby video.