Existing RGB-T semantic segmentation networks adopt the one-way RGB-T data fusion strategy that enhances RGB features with thermal features, resulting in the underutilization of thermal information. To address it, we propose a novel RGB-T fusion network termed MMNet which introduces an adaptively mutual enhancement mechanism to enhance RGB and thermal features by each other in the encoder, as well as hierarchically aggregates multi-resolution RGB-T features in the decoder. Experimental results show that MMNet achieves state-of-the-art 56.8% mIoU, being 1.66× faster than the best previous work.
MMNet: RGB-t Semantic Segmentation Network Based on Multi-scale and Adaptively Mutual Enhancement Mechanism
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2021 ; Changsha, China September 24, 2021 - September 26, 2021
Proceedings of 2021 International Conference on Autonomous Unmanned Systems (ICAUS 2021) ; Chapter : 338 ; 3435-3444
2022-03-18
10 pages
Article/Chapter (Book)
Electronic Resource
English