Provided are a face beauty prediction method based on multi-task migration and a device, the method comprises: carrying out similarity measurement of a plurality of tasks based on a graph structure, to obtain optimal combination of the plurality of tasks (S100); constructing a face beauty prediction model comprising a feature sharing layer based on the optimal combination (S200); migrating feature parameters of a prior large-scale face image network to the feature sharing layer of the face beauty prediction model (S300); inputting face images for training to pre-train the face beauty prediction model (S400); and inputting a face image to be tested into the trained face beauty prediction model to obtain a face recognition result (S500). The effects of reducing the redundancy of the deep learning task, reducing the burden of network training, and improving the efficiency and precision of network classification and recognition are achieved.
展开▼