Multi-object image parsing using neural network pipeline

A system to modify a segment of a human in a source image that includes at least the human, the system comprising:

• one or more processors;
• an image parsing pipeline executable by the one or more processors, the image parsing pipeline comprising
• an object detection module to detect the human in the source image,
• a segmentation module to generate a preliminary mask image that identifies a plurality of segments of the human in the source image, and
• a refinement module to refine the identification of the plurality of segments of the human in the preliminary mask image, and to generate a final mask image based on the refinement;
• and
• an image modification module executable by the one or more processors to modify a segment of the human in the source image, based on the final mask image identifying pixels of the segment to be modified.

A computer program product including one or more non-transitory machine-readable mediums encoded with instructions that when executed by one or more processors cause a process to be carried out for identifying segments of an object in a source image, the process comprising:

• detecting, by a first neural network model, the object within the source image;
• generating, by a second neural network model, a preliminary mask image that identifies a plurality of segments of the object;
• and
• generating, by a third neural network model, a final mask image, by refining the identification of the plurality of segments of the object in the preliminary mask image.