Seed, Expand and Constrain: Three Principles for Weakly-Supervised Image Segmentation

TL;DR

This paper proposes a new loss function for weakly-supervised image segmentation that improves accuracy by guiding seed initialization, expansion based on class information, and boundary constraints, outperforming previous methods.

Contribution

Introduction of a novel loss function based on three principles for weakly-supervised segmentation, leading to significant performance improvements.

Findings

Outperforms previous state-of-the-art on PASCAL VOC 2012

Detailed analysis of each loss component's impact on segmentation quality

Effective integration of seed, expand, and constrain principles

Abstract

We introduce a new loss function for the weakly-supervised training of semantic image segmentation models based on three guiding principles: to seed with weak localization cues, to expand objects based on the information about which classes can occur in an image, and to constrain the segmentations to coincide with object boundaries. We show experimentally that training a deep convolutional neural network using the proposed loss function leads to substantially better segmentations than previous state-of-the-art methods on the challenging PASCAL VOC 2012 dataset. We furthermore give insight into the working mechanism of our method by a detailed experimental study that illustrates how the segmentation quality is affected by each term of the proposed loss function as well as their combinations.