We present a novel Transformer network, called LKFormer, to solve infrared image SR. Specifically, we design a large kernel residual attention structure to replace the vanilla SA layer to achieve local and non-local feature modeling. In addition, the proposed module can process high-resolution infrared images more efficiently and does not exhibit a quadratic increase in computational complexity with increasing image resolution. Furthermore, to enhance the suitability of the proposed Transformer architecture for the task of dense pixel prediction, we develop a novel module, named the GPFN. The GPFN module improves the information flow within the network by incorporating pixel attention branching. The link to the paper is at LKFormer: large kernel transformer for infrared image super-resolution | Multimedia Tools and Applications (springer.com).
- Please prepare an environment with python=3.8, and then use the pyTorch.
- First configure the parameters under the option file and then run the main_train_SR file.
We use public datasets.
Donwload link is available at https://figshare.com/s/2121562561211c0a8101 .