A Variant of WSL Framework For Weakly Supervised Semantic Segmentation

Abstract

Scene understanding is an important task in the field of machine vision. Image semantic segmentation is helpful to realizes scene understanding by identifying semantic information in images. Due to fully supervised semantic segmentation need a lot of manual annotation, but its costs is so expensive. So Weakly supervised Learning(WSL) become more and more popular. In this paper, we analyze this problem from a new perspective. We only use a small amount of data as a training set and solve the problem with only the image label. First, we only train the existing neural network to complete the weakly supervised semantic segmentation task; Second, we change classification network from a WSL framework to encourage neural networks to identify semantic information in images. We conducted experiments on the PASCAL VOC 2012 dataset, and our results have some improvement in semantic segmentation.