Apparatus and method for reconstructing 3D human object based on monocular image with depth image-based implicit function learning

A three-dimensional (3D) human object reconstruction apparatus comprising:

• one or more processors comprising:
• a first prediction unit configured to predict an orthographic image from a front perspective color image of a human object, and to predict a depth map from the predicted orthographic image;
• and
• a second prediction unit configured to predict a signed distance (SD) regarding points on a 3D space from the predicted depth map,
• wherein the one or more processors are further configured to:
• extract a pixel-aligned feature from the predicted depth map;
• indicate points on the 3D space one by one in sequence;
• extract a ray-invariant feature from the pixel-aligned feature;
• generate encoded position information by encoding position information of the indicated points;
• and
• predict a SD of a point from the extracted ray-invariant feature and the encoded position information.

A three-dimensional (3D) human object reconstruction method, performed by one or more processors, the method comprising:

• predicting a double-sided orthographic depth map from a front perspective color image of a human object;
• predicting a signed distance (SD) regarding points on a 3D space from the predicted double-sided orthographic depth map;
• and
• reconstructing a 3D human object by using the predicted SD,
• wherein the predicting the signed distance (SD) comprises:
• extracting a pixel-aligned feature from the predicted double-sided orthographic depth map;
• indicating points on the 3D space one by one in sequence;
• extracting a ray-invariant feature from the pixel-aligned feature;
• generating encoded position information by encoding position information of the indicated points;
• and
• predicting a SD of a point from the extracted ray-invariant feature and the encoded position information.