Lightweight single-image super-resolution network based on dual paths
Li Ke, Liu Yukai
TL;DR
The paper addresses the challenge of efficient single-image super-resolution by balancing local detail capture and global context under limited parameter budgets. It introduces a lightweight dual-path network that fuses Transformer-based local features with a convolutional branch that yields global coarse information, leveraging a two-way interaction and multi-stage feature supplementation to preserve restoration-relevant cues. Key contributions include (1) a novel two-way fusion network combining Axis Transformer Block and Space Enhanced Self-Attention Block, (2) a compact convolutional branch with depth-separable convolutions and CA/ESA modules, (3) the multi-scale dual-path feature fusion block and a Feature Reuse Block to integrate multi-depth features, and (4) empirical evidence that the approach achieves state-of-the-art performance among lightweight SISR methods on standard benchmarks. Experiments show improved PSNR/SSIM and richer texture reconstruction compared with contemporary lightweight rivals, supporting its suitability for deployment on resource-constrained devices.
Abstract
The single image super-resolution(SISR) algorithms under deep learning currently have two main models, one based on convolutional neural networks and the other based on Transformer. The former uses the stacking of convolutional layers with different convolutional kernel sizes to design the model, which enables the model to better extract the local features of the image; the latter uses the self-attention mechanism to design the model, which allows the model to establish long-distance dependencies between image pixel points through the self-attention mechanism and then better extract the global features of the image. However, both of the above methods face their problems. Based on this, this paper proposes a new lightweight multi-scale feature fusion network model based on two-way complementary convolutional and Transformer, which integrates the respective features of Transformer and convolutional neural networks through a two-branch network architecture, to realize the mutual fusion of global and local information. Meanwhile, considering the partial loss of information caused by the low-pixel images trained by the deep neural network, this paper designs a modular connection method of multi-stage feature supplementation to fuse the feature maps extracted from the shallow stage of the model with those extracted from the deep stage of the model, to minimize the loss of the information in the feature images that is beneficial to the image restoration as much as possible, to facilitate the obtaining of a higher-quality restored image. The practical results finally show that the model proposed in this paper is optimal in image recovery performance when compared with other lightweight models with the same amount of parameters.