UCAN

<p align="center"> <h1 align="center"> <img width="200" src="logo.png"> <br> UCAN: Unified Convolutional Attention Network <br> for Expansive Receptive Fields in Lightweight Super-Resolution</h1> <p align="center"> Cao Thien Tan^1,2,3 · Phan Thi Thu Trang ^3,5 · Do Nghiem Duc</a>⁶ · Ho Ngoc Anh</a>⁵ · Hanyang Zhuang</a>⁴ . <br> Nguyen Duc Dung</a>² </p> <p align="center"> ¹Ho Chi Minh City Open University &nbsp; &nbsp; ²AI Tech Lab, Ho Chi Minh City University of Technology <br> ³Code Mely AI Research Team &nbsp; &nbsp; ⁴Global College, Shanghai Jiao Tong University <br> ⁵Ha Noi University of Science and Technology &nbsp; &nbsp; ⁶University of Manitoba </p> <h3 align="center"> CVPR 2026 - Poster </h3> <h3 align="center"><a href="https://arxiv.org/abs/2603.11680">[Paper]</a></h3> </p>

---

Abstract: Hybrid CNN-Transformer architectures achieve strong results in image super-resolution, but scaling attention windows or convolution kernels significantly increases computational cost, limiting deployment on resource-constrained devices. We present UCAN, a lightweight network that unifies convolution and attention to expand the effective receptive field efficiently. UCAN combines window-based spatial attention with a Hedgehog Attention mechanism to model both local texture and long-range dependencies, and introduces a distillation-based large-kernel module to preserve high-frequency structure without heavy computation. In addition, we employ cross-layer parameter sharing to further reduce complexity. On Manga109 (4×), UCAN-L achieves 31.63 dB PSNR with only 48.4G MACs, surpassing recent lightweight models. On BSDS100, UCAN attains 27.79 dB, outperforming methods with significantly larger models. Extensive experiments show that UCAN achieves a superior trade-off between accuracy, efficiency, and scalability, making it well-suited for practical high-resolution image restoration.

Paper will be updated soon.

🔥 News

• 2026-02: 🎉UCAN is accepted by CVPR 2026! This repo is released.

🛠️ Setup

🗃️ Datasets

The datasets used in our training and testing are orgnized as follows:

Training and testing sets can be downloaded as follows:

| Training Set | Testing Set | Visual Results | | :-----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: | | DIV2K (800 training images, 100 validation images) [organized training dataset DIV2K: One Drive] | Set5 + Set14 + BSD100 + Urban100 + Manga109 [complete testing dataset: One Drive] | Updating |

💡 Models

🚀 Training

🧪 Testing

📚 Results

🙇 Citation

🥂 Acknowledgements

This work is based on BasicSR, HiT-SR, ESC and MambaIRv2. We thank them for their great work and for sharing the code.