Efficient Single Image Super-Resolution with Entropy Attention and Receptive Field Augmentation

we present an efficient SR model to mitigate the dilemma between model efficiency and SR per

formance, which is dubbed Entropy Attention and Receptive Field Augmentation network (EARFA), and composed of a novel entropy attention (EA) and a shifting large kernel attention (SLKA).

From the perspective of information theory , EA is introduced into the model to elevate the entropy of intermediate features conditioned on a Gaussian distribution, and thus increase the input information for subsequent inference. Specifically, it computes the differential entropy [ 32] for channel-wise features, which is used to measure the information amount in randomly distributed data. And the attention weights are obtained by driving the features approaching to a Gaussian distribution.

[32] Claude Elwood Shannon. 1948. A mathematical theory of communication. The

Bell System Technical Journal 27, 3 (1948), 379–423

SLKA is an improved version of lager kernel attention (LKA) [ 10 ] aimed at further augmenting the

effective receptive field of the model with negligible overhead. This is implemented by simply shifting partial channels of a intermediate feature [42].

[10] Meng-Hao Guo, Cheng-Ze Lu, Zheng-Ning Liu, Ming-Ming Cheng, and Shi-Min

Hu. 2023. Visual attention network. Computational Visual Media 9, 4 (2023),

733–752.

[42] Xiaoming Zhang, Tianrui Li, and Xiaole Zhao. 2023. Boosting Single Image

Super-Resolution via Partial Channel Shifting. In Proceedings of the IEEE/CVF

International Conference on Computer Vision . IEEE, 13223–13232.