Among many semantic segmentation works using deep learning methods, fusing multiple layer features usually could boost performance. Multi-layer feature fusion could obtain more comprehensive context information. However, fusing different layers leads to different experiment results. There is no unified method to select effective layers to fuse in previous works, which mostly relied on intuition or experience. In this paper, we propose a fusion coefficient learning method that can guide us to select effective layers. What’s more, our approaches can be added to other works that require multi-scale fusion to further boost their performance. We proposed three principles for preliminary screening of layers and presented the fusion coefficient learning algorithm. Then, We could select the most effective layer through three steps. Our approaches are verified by massive experiments and proved to be effective on the PASCAL VOC2012, PASCAL Context data set.
CITATION STYLE
Yu, J., Zhao, S., & Han, Y. (2018). Choose the largest contributor: A fusion coefficient learning network for semantic segmentation. In Communications in Computer and Information Science (Vol. 819, pp. 54–64). Springer Verlag. https://doi.org/10.1007/978-981-10-8530-7_6
Mendeley helps you to discover research relevant for your work.