Reconstructing three-dimensional models of objects from real images based on depth information

A system comprising:

• one or more memory devices comprising real two-dimensional (2D) images and a three-dimensional (3D)-object-reconstruction-machine-learning model;
• and
• one or more processors configured to cause the system to:
• determine a predicted depth map and a segmentation mask for a real 2D image depicting a digital object;
• generate, utilizing the 3D-object-reconstruction-machine-learning model, a predicted signed distance function based on the predicted depth map and the segmentation mask;
• pull a query point in a spatial representation of the real 2D image towards a surface depth point in the spatial representation by moving the query point to a pulled point according to the predicted signed distance function;
• determine a depth differential between the pulled point and the surface depth point in the spatial representation;
• and
• modify one or more learned parameters of the 3D-object-reconstruction-machine-learning model based on the depth differential.

A computer-implemented method comprising:

• generating a depth map and a segmentation mask for a real two-dimensional (2D) image depicting a digital object;
• utilizing a three-dimensional (3D)-object-reconstruction-machine-learning model to generate a predicted signed distance function based on the depth map and the segmentation mask;
• performing a step for determining a depth differential among points in a spatial representation of the real 2D image according to the predicted signed distance function;
• and
• modifying one or more learned parameters of the 3D-object-reconstruction-machine-learning model according to the depth differential.