Importance Sampling CAMs for Weakly-Supervised Segmentation

TL;DR

This paper introduces importance sampling and a feature similarity loss to improve class activation maps for weakly-supervised segmentation, resulting in more accurate and well-defined object contours without pixel-level annotations.

Contribution

It proposes a novel importance sampling approach and a feature similarity loss to enhance CAMs for weakly-supervised segmentation, addressing focus and contour issues.

Findings

Significant improvement in contour accuracy on PASCAL VOC 2012

Comparable region similarity to state-of-the-art methods

Enhanced CAM activation over larger object extents

Abstract

Classification networks can be used to localize and segment objects in images by means of class activation maps (CAMs). However, without pixel-level annotations, classification networks are known to (1) mainly focus on discriminative regions, and (2) to produce diffuse CAMs without well-defined prediction contours. In this work, we approach both problems with two contributions for improving CAM learning. First, we incorporate importance sampling based on the class-wise probability mass function induced by the CAMs to produce stochastic image-level class predictions. This results in CAMs which activate over a larger extent of objects. Second, we formulate a feature similarity loss term which aims to match the prediction contours with edges in the image. As a third contribution, we conduct experiments on the PASCAL VOC 2012 benchmark dataset to demonstrate that these modifications significantly increase the performance in terms of contour accuracy, while being comparable to current state-of-the-art methods in terms of region similarity.