Research on High-Precision Cloth Animation Synthesis Method Based on Deep Learning

Aiming at the shortcomings of long time-consuming and poor real-time performance in cloth simulation method based on physical model, a high-precision cloth animation synthesis method based on deep learning is proposed. Firstly, high-precision cloth animation is obtained by physical method simulation; Then each frame grid is converted into a “geometric image” (RGB image storing vertex displacement), and then down-sampled into a low-resolution geometric image. Using these image pairs, an image super-resolution network is trained, which takes $2\mathrm{N}+1$ continuous low-resolution geometric images as input to predict the central high-resolution geometric image. Finally, the predicted high-resolution geometric image is reconstructed to obtain a three-dimensional mesh model.