Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RT-Focuser: A Real-Time Lightweight Model for Edge-side Image Deblurring

Zhuoyu Wu, Wenhui Ou, Qiawei Zheng, Jiayan Yang, Quanjun Wang, Wenqi Fang, Zheng Wang, Yongkui Yang, Heshan Li

TL;DR

The paper tackles real-time image deblurring on edge devices by introducing RT-Focuser, a compact SISO U-shaped network. It leverages three novel modules—Lightweight Deblurring Block (LD), Multi-Level Integrated Aggregation (MLIA), and Cross-source Fusion Block (X-Fuse)—to preserve edges, integrate multi-scale encoder features, and progressively refine the decoder. RT-Focuser achieves a highly favorable speed-to-quality balance (5.85M parameters, 15.76 GMACs, ~6 ms per frame, >140 FPS on mobile/GPU) with PSNR of 30.67 dB on the GoPro dataset, outperforming many prior lightweight methods in efficiency while maintaining competitive restoration quality. The results demonstrate strong potential for real-time, edge-side deployment in applications like autonomous driving, UAV perception, and medical imaging.

Abstract

Motion blur caused by camera or object movement severely degrades image quality and poses challenges for real-time applications such as autonomous driving, UAV perception, and medical imaging. In this paper, a lightweight U-shaped network tailored for real-time deblurring is presented and named RT-Focuser. To balance speed and accuracy, we design three key components: Lightweight Deblurring Block (LD) for edge-aware feature extraction, Multi-Level Integrated Aggregation module (MLIA) for encoder integration, and Cross-source Fusion Block (X-Fuse) for progressive decoder refinement. Trained on a single blurred input, RT-Focuser achieves 30.67 dB PSNR with only 5.85M parameters and 15.76 GMACs. It runs 6ms per frame on GPU and mobile, exceeds 140 FPS on both, showing strong potential for deployment on the edge. The official code and usage are available on: https://github.com/ReaganWu/RT-Focuser.

RT-Focuser: A Real-Time Lightweight Model for Edge-side Image Deblurring

TL;DR

The paper tackles real-time image deblurring on edge devices by introducing RT-Focuser, a compact SISO U-shaped network. It leverages three novel modules—Lightweight Deblurring Block (LD), Multi-Level Integrated Aggregation (MLIA), and Cross-source Fusion Block (X-Fuse)—to preserve edges, integrate multi-scale encoder features, and progressively refine the decoder. RT-Focuser achieves a highly favorable speed-to-quality balance (5.85M parameters, 15.76 GMACs, ~6 ms per frame, >140 FPS on mobile/GPU) with PSNR of 30.67 dB on the GoPro dataset, outperforming many prior lightweight methods in efficiency while maintaining competitive restoration quality. The results demonstrate strong potential for real-time, edge-side deployment in applications like autonomous driving, UAV perception, and medical imaging.

Abstract

Motion blur caused by camera or object movement severely degrades image quality and poses challenges for real-time applications such as autonomous driving, UAV perception, and medical imaging. In this paper, a lightweight U-shaped network tailored for real-time deblurring is presented and named RT-Focuser. To balance speed and accuracy, we design three key components: Lightweight Deblurring Block (LD) for edge-aware feature extraction, Multi-Level Integrated Aggregation module (MLIA) for encoder integration, and Cross-source Fusion Block (X-Fuse) for progressive decoder refinement. Trained on a single blurred input, RT-Focuser achieves 30.67 dB PSNR with only 5.85M parameters and 15.76 GMACs. It runs 6ms per frame on GPU and mobile, exceeds 140 FPS on both, showing strong potential for deployment on the edge. The official code and usage are available on: https://github.com/ReaganWu/RT-Focuser.
Paper Structure (9 sections, 2 figures, 2 tables)

This paper contains 9 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Lightweight Deblurring Block (LD)
  4. Cross-source Fusion Block (X-Fuse)
  5. Multi-Level Integrated Aggregation Module (MLIA)
  6. Experiments
  7. Comparison with Advanced Models
  8. Deployment Efficiency on Edge and General Platforms
  9. Conclusion

Figures (2)

  • Figure 1: Visual and performance comparison. RT-Focuser shows clear visual recovery (left) and achieves strong trade-offs in PSNR, SSIM, and efficiency metrics (right).
  • Figure 2: Overview of RT-Focuser. It follows a U-shaped encoder–decoder structure with multi-scale fusion. Blue and green arrows indicate 3×3 convolution and ×2 bilinear upsampling, respectively. Down (2X) is the bilinear interpolation for downsampling.