Method and system for instance segmentation of densely packed objects

A system for instance segmentation of densely packed objects, the system comprising:

• a processor and a memory communicatively coupled to the processor, wherein the memory stores processor-executable instructions, which, on execution, causes the processor to:
• receive an input image comprising a plurality of densely packed objects;
• enhance the input image to generate an enhanced input image;
• input the enhanced input image to a trained mask region-based convolutional neural network (R-CNN) model to generate a first segmentation map, wherein the first segmentation map comprises a first set of instances of densely packed objects;
• iteratively predict one or more missing instances of densely packed objects from the input image based on comparison of a contour map generated for the input image with a contour map generated for the first segmentation map;
• compare an area of the one or more missing instances with a predefined threshold value, wherein iterative prediction of the one or more missing instances is performed till value of the area is above the predefined threshold value;
• apply, at one or more iterations based on comparison of the area with the predefined threshold value, the trained masked R-CNN model on each of the one or more missing instances to generate a second segmentation map, wherein the second segmentation map comprises the one or more missing instances;
• and
• obtain, at each of the one or more iterations, a refined segmentation map based on the first segmentation map and the second segmentation map, wherein the refined segmentation map comprises the first set of instances and the one or more missing instances, and wherein the refined segmentation map corresponds to input image for a succeeding iteration, till the area of missing instances in a current iteration is above the predefined threshold value.

A computer program product for instance segmentation of densely packed objects, the computer program product embodied in a non-transitory computer readable storage medium and comprising computer-executable instructions configured for:

• receiving an input image comprising a plurality of densely packed objects;
• enhancing the input image to generate an enhanced input image;
• inputting the enhanced input image to a trained mask region-based convolutional neural network (R-CNN) model to generate a first segmentation map, wherein the first segmentation map comprises a first set of instances of densely packed objects;
• iteratively predicting one or more missing instances of densely packed objects from the input image based on comparison of a contour map generated for the input image with a contour map generated for the first segmentation map;
• comparing an area of the one or more missing instances with a predefined threshold value, wherein iterative prediction of the one or more missing instances is performed till value of the area is above the predefined threshold value;
• applying, at one or more iterations based on comparison of the area with the predefined threshold value, the trained masked R-CNN model on each of the one or more missing instances to generate a second segmentation map, wherein the second segmentation map comprises the one or more missing instances;
• and
• obtaining, at each of the one or more iterations, a refined segmentation map based on the first segmentation map and the second segmentation map, wherein the refined segmentation map comprises the first set of instances and the one or more missing instances, and wherein the refined segmentation map corresponds to input image for a succeeding iteration, till the area of missing instances in a current iteration is above the predefined threshold value.