UCAN: Unified Convolutional Attention Network
for Expansive Receptive Fields in Lightweight Super-Resolution

Cao Thien Tan^1,2,3 · Phan Thi Thu Trang ^3,5 · Do Nghiem Duc⁶ · Ho Ngoc Anh⁵ · Hanyang Zhuang⁴ .
Nguyen Duc Dung²

¹Ho Chi Minh City Open University     ²AI Tech Lab, Ho Chi Minh City University of Technology
³Code Mely AI Research Team     ⁴Global College, Shanghai Jiao Tong University
⁵Ha Noi University of Science and Technology     ⁶University of Manitoba

CVPR 2026 - Poster

[Paper]

---

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.

Repo updating...

🔥 News

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

🛠️ Setup

• Python 3.10+
• PyTorch 2.0+ + Torchvision 0.15+
• NVIDIA GPU + CUDA

git clone https://github.com/hokiyoshi/UCAN.git
conda create -n UCAN python=3.10
conda activate UCAN
pip install -r requirements.txt
python setup.py develop

🗃️ Datasets

Training and testing sets can be downloaded as follows:

Training Set	Testing Set	Visual Results	Log	Weight
DIV2K (800 training images, 100 validation images) [organized training dataset DIV2K: One Drive]	Set5 + Set14 + BSD100 + Urban100 + Manga109 [complete testing dataset: One Drive]	Updating	x2, x3, x4	Link

💡 Models

🚀 Training

python basicsr/train.py -opt options/Train/train_UCAN_x2.yml

🧪 Testing

Place pretrained weights under experiments/pretrained_models/, then run:

# x2
python basicsr/test.py -opt options/Test/test_UCAN_x2.yml
# x3
python basicsr/test.py -opt options/Test/test_UCAN_x3.yml
# x4
python basicsr/test.py -opt options/Test/test_UCAN_x4.yml

Results are saved to results/.

📚 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.