Image segmentation with object boundary refinement

A system, comprising:

• a memory that stores instructions, and
• a processor that is coupled to the memory and, when executing the instructions, is configured to:
• receive an image and an indication of a region within the image, wherein the region identifies an outermost portion of the image where pixels of an object are likely to occur, wherein the indication comprises a bounding box drawn by a user via a graphical user interface;
• determine, by a machine learning model, a per-pixel score for each pixel within the region, wherein, for each pixel, the per-pixel score indicates a likelihood that the pixel is included in the object;
• determine a boundary of the object within the region based on the per-pixel scores for pixels within the region;
• receive a boundary parameter that indicates a score threshold to be applied to per-pixel scores of the pixels of the region;
• determine a portion of the region that includes the object by applying the score threshold to the per-pixel scores of the pixels, wherein only pixels having per-pixel scores which exceed the score threshold are included in the portion;
• receive a second boundary parameter that indicates a second score threshold to be applied to per-pixel scores of the pixels of the region, wherein the second score threshold is different from the score threshold;
• determine a second portion of the region that includes the object by applying the second score threshold to the per-pixel scores of the pixels, wherein only pixels having per-pixel scores which exceed the second score threshold are included in the second portion;
• and
• via the graphical user interface, display the portion upon selection of the boundary parameter by the user and display the second portion upon selection of the second boundary parameter by the user.

One or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of:

• receiving an image and an indication of a region within the image, wherein the region identifies an outermost portion of the image where pixels of an object are likely to occur, wherein the indication comprises a bounding box drawn by a user via a graphical user interface;
• determining, by a machine learning model, a per-pixel score for each pixel within the region, wherein, for each pixel, the per-pixel score indicates a likelihood that the pixel is included in the object;
• determining a boundary of the object within the region based on the per-pixel scores for pixels within the region;
• receiving a boundary parameter that indicates a score threshold to be applied to per-pixel scores of the pixels of the region;
• determining a portion of the region that includes the object by applying the score threshold to the per-pixel scores of the pixels, wherein only pixels having per-pixel scores which exceed the score threshold are included in the portion;
• receiving a second boundary parameter that indicates a second score threshold to be applied to per-pixel scores of the pixels of the region, wherein the second score threshold is different from the score threshold;
• determining a second portion of the region that includes the object by applying the second score threshold to the per-pixel scores of the pixels, wherein only pixels having per-pixel scores which exceed the second score threshold are included in the second portion;
• and
• via the graphical user interface, displaying the portion upon selection of the boundary parameter by the user and displaying the second portion upon selection of the second boundary parameter by the user.