In deep learning-based methods, many approaches, such as network architecture modification and input source diversification, have been applied to achieve performance improvement. However, existing methods have developed in their own aspect rather than taking advantage of other methods. Furthermore, to the best of our knowledge, no attempt has been made to merge the well performing networks. In this paper, we proposed a network that improves performance and reliability by using state-of-the-art networks as the baseline and fusing the results obtained by the baseline network. We validate the efficacy of our fusion framework on the task of semantic segmentation; we compare the results from SOTA methods with that of our SOTA-fusion framework.