Abstract: Methods, mediums, and systems for constructing a 3D face model, including Deep Iterative Face Fitting (DIFF) and Recurrent Feature Alignment (ReFA). A processor provides a reference—containing a 3D mesh representing a median face, vertex positions, or surface points—as a template and UV texture mapping pointing the 3D mesh to a 2D UV space; receives input image(s) of a face and extracts geometry/texture features in an image space; extracts features in a UV space; iteratively produces a feature map via visual semantic correlation between UV and image spaces and regress geometry updates, predicting texture maps and comparing features, and inputs the map to an RNN-based neural optimizer of Gated Recurrent Units (GRU) to determine a hidden state. A head pose and/or updated 3D mesh/UV-space position map is output, each pixel in the UV-space map storing a coordinate of a corresponding point in a canonical space of the 3D mesh.

Abstract: This application relates to a power consumption grading method for a network-based power supply system. The method includes: when it is detected that power overload protection is not triggered, starting a plurality of power modules of a powered device one by one in a specific order until all the plurality of power modules are started or it is detected that the power overload protection is triggered after a specific power module is started; when it is detected that the power overload protection is triggered, selecting, according to the specific order and based on the specific power module, one or more of the plurality of power modules as a power module combination of the powered device; and determining a load power of the powered device and a corresponding power consumption level based on the power module combination.