Distance to Center of Mass Encoding for Instance Segmentation

TL;DR

This paper introduces DCME, a novel encoding method for instance segmentation that represents objects by their center of mass and a vector field, enabling models to learn precise object delineation.

Contribution

The paper proposes a new mathematical encoding for instance segmentation based on center of mass and vector fields, facilitating improved learning and generalization.

Findings

Enables precise object boundary delineation.

Facilitates partial occlusion evaluation.

Improves model generalization for segmentation.

Abstract

The instance segmentation can be considered an extension of the object detection problem where bounding boxes are replaced by object contours. Strictly speaking the problem requires to identify each pixel instance and class independently of the artifice used for this mean. The advantage of instance segmentation over the usual object detection lies in the precise delineation of objects improving object localization. Additionally, object contours allow the evaluation of partial occlusion with basic image processing algorithms. This work approaches the instance segmentation problem as an annotation problem and presents a novel technique to encode and decode ground truth annotations. We propose a mathematical representation of instances that any deep semantic segmentation model can learn and generalize. Each individual instance is represented by a center of mass and a field of vectors pointing to it. This encoding technique has been denominated Distance to Center of Mass Encoding (DCME).