Contour Loss for Instance Segmentation via k-step Distance Transformation Image

TL;DR

This paper introduces a novel contour loss function for instance segmentation that enhances mask accuracy near object boundaries by focusing on contour optimization using a differentiable distance transformation module, improving performance on COCO.

Contribution

The paper proposes a new contour loss and a differentiable k-step distance transformation module to improve boundary accuracy in instance segmentation models like Mask R-CNN.

Findings

Contour loss improves segmentation accuracy near object contours.

The method is compatible with existing models like Mask R-CNN.

Experimental results show enhanced performance on COCO dataset.

Abstract

Instance segmentation aims to locate targets in the image and segment each target area at pixel level, which is one of the most important tasks in computer vision. Mask R-CNN is a classic method of instance segmentation, but we find that its predicted masks are unclear and inaccurate near contours. To cope with this problem, we draw on the idea of contour matching based on distance transformation image and propose a novel loss function, called contour loss. Contour loss is designed to specifically optimize the contour parts of the predicted masks, thus can assure more accurate instance segmentation. In order to make the proposed contour loss to be jointly trained under modern neural network frameworks, we design a differentiable k-step distance transformation image calculation module, which can approximately compute truncated distance transformation images of the predicted mask and corresponding ground-truth mask online. The proposed contour loss can be integrated into existing instance segmentation methods such as Mask R-CNN, and combined with their original loss functions without modification of the inference network structures, thus has strong versatility. Experimental results on COCO show that contour loss is effective, which can further improve instance segmentation performances.