3D Face Reconstruction Algorithm Based on Lightweight Attention Mechanism

3D face reconstruction has significant research value in computer vision sector. This technic was widely applied in multiple areas, such as virtual try-on, E-commerce live streaming, anti-proofing and animation. But reconstructing by traditional algorithms may affected by facial expression, occlusion and ambient light that can cause poor reconstruction accuracy and robustness. As artificial intelligence developing fast, deep learning was more and more used in computer vision sector. That provides various possibilities in optimizing of 3D face reconstruction algorithms.. We developed a three-dimensional face reconstruction algorithm based on light attention proposed in this thesis. The algorithm adds face mask input and channel attention mechanism to the preprocessing part and the network structure part respectively, and uses a lightweight module in the network structure part. The algorithm greatly reduces the number of parameters of the algorithm model, improves the reconstruction speed and reconstruction effect, and has strong practicability.