Semantic segmentation, a pixel-level classification task, is crucial for the fine-grained classification of objects within images. However, its reliance on precise pixel-level labeling poses a significant challenge, increasing costs and limiting its applicability in real-world scenarios. Despite the semi-supervised learning methods that have alleviated the need for extensive labeled data, many still involve complex processes or substantial additional resources. We propose a similarity-based training strategy and a simple model configured with the online and the target network to perform semi-supervised semantic segmentation while reducing the required resources and maintaining a simpler configuration than conventional methods. To assess the effectiveness of our method, we conducted evaluations using various splits of the PASCAL VOC 2012 dataset, comparing it with other semi-supervised semantic segmentation approaches. Experimental results demonstrate that our proposed method outperforms conventional methods that rely on intricate processes or additional computational resources. This suggests the potential for a more practical and resource-efficient approach to semi-supervised semantic segmentation tasks.