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 disclosure provides techniques of implementing online somatosensory dance competitions. The techniques comprise assigning a room to one or more players in response to receiving one or more requests of starting a game sent by the one or more players, wherein all players in a same room share a same dance video; outputting a plurality of standard movement images corresponding to the same dance video in a chronological order, wherein the plurality of standard movement images each indicate position coordinates of different skeleton nodes at a corresponding moment; receiving dance movement signals associated with each of the one or more players, and determining scores for each player based on a similarity between the dance movement signals and standard movement parameters, wherein the standard movement parameters correspond to the standard movement images; and collecting and outputting statistics on an overall score and ranking of each of the one or more players.

Abstract: This disclosure provides techniques of extracting movements from a dance video. The techniques comprise receiving a dance video that includes one or more dancers and that is uploaded by a user, and obtaining a video frame in the dance video; recognizing, based on a skeleton node recognition algorithm, skeleton node images from the video frames corresponding to a target dancer, where the target dancer is selected from the one or more dancers; performing cluster analysis on the skeleton node images recognized from the dance video and corresponding to each target dancer to obtain a plurality of cluster sets; determining a cluster center in each of the plurality of cluster sets as a key skeleton node image; and sequentially outputting the key skeleton node images to obtain a standard movement sequence corresponding to each target dancer in the dance video.