Lightweight AI Network Reconstructs High-Resolution Images Under One Million Parameters

Researchers have introduced a new deep learning model designed to sharpen low-resolution images into high-resolution ones while keeping the model itself remarkably small. The method, called a hierarchical feature attention network (HFAN), is described in a paper published in Cluster Computing, and its authors report that the entire network stays below one million parameters.

Image super-resolution, the task of reconstructing high-resolution images from their low-resolution counterparts, is a central problem in computer vision. According to the paper’s abstract, deep learning approaches have made significant progress in this area, but existing methods often struggle to balance reconstruction quality against computational complexity. The new network is presented as a step toward addressing that trade-off.

At the core of the proposed system is a hierarchical feature attention module (HFAM). The module is designed to capture effective long-range dependencies by integrating multi-level feature information. It does so by constructing parallel global and local feature extraction branches, which the authors say allow the model to simultaneously capture long-range dependencies and fine local details, and to fuse global context with local information effectively.

The architecture also includes a Feature Enhancement Module, which the researchers introduce to integrate spatial information for high-quality reconstruction of fine-grained textures.

The team reports extensive experiments on multiple benchmark datasets. According to the abstract, the proposed method achieves highly competitive performance compared with state-of-the-art lightweight methods in terms of both objective metrics and visual perceptual quality. The published material does not report specific numerical scores, so the strength of the comparison rests on the authors’ summary of their experiments.

The work is authored by Dong Li, Silu Yang and Zhaohong Liu, with Heng Zhang also listed among the authors, and appears in Volume 29 of Cluster Computing as article number 872. The paper was published on 22 October 2026.

As with any newly published method, the reported performance reflects the authors’ own benchmark experiments as summarized in the abstract, and independent evaluation of the network’s behavior across different image types and deployment conditions would come from future work beyond the scope of the published preview.

Subject of Research: A lightweight deep learning network for image super-resolution

Article Title: Hierarchical feature attention network for image super-resolution

Article References: Li, D., Yang, S., Liu, Z., & Zhang, H. (2026). Hierarchical feature attention network for image super-resolution. Cluster Computing, 29(15), Article 872. https://doi.org/10.1007/s10586-026-06707-4

Image Credits: AI Generated

DOI: 10.1007/s10586-026-06707-4

Keywords: image super-resolution, deep learning, computer vision, hierarchical feature attention, lightweight networks, feature extraction, image reconstruction, neural networks, benchmark datasets, Cluster Computing, Hierarchical, feature