University of Manchester Develops Sub-1K Parameter AI for Low-Light Enhancement
Researchers at the University of Manchester have developed Multinex, an ultra-lightweight AI model for low-light image enhancement that combines Retinex theory with neural networks. This innovation significantly reduces computational costs, allowing deployment on edge devices like smartphones and security cameras, and surpasses more cumbersome models in real-time performance. The research was validated and presented at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026.
Key Takeaways
- Multinex, developed by Alexandru Brateanu, combines classical Retinex theory with neural networks for low-light image enhancement.
- The lightweight version uses 45,000 parameters, and the nano variant operates with 700 parameters.
- Multinex achieved real-time performance and surpassed models like PairLIE (330K parameters) and ZeroDCE (80K parameters).
- The model is suitable for edge devices such as smartphones, autonomous robots, and security cameras due to its low computational cost.
Why It Matters
This development in low-light AI means enhanced visual clarity on resource-constrained devices, impacting remote monitoring and autonomous operations. The blend of classical theory with neural networks could drive more efficient AI solutions across various computer vision tasks. Watch for integration into commercial camera systems and its impact on real-world AI applications needing robust perception in challenging conditions.
Additional Context
Following the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026 presentation, Multinex’s official implementation, including training and testing code, has been made available via a GitHub repository (albrateanu/multinex, March 2026). This includes pretrained checkpoints for direct evaluation and supports Pytorch 2 environments. The availability of Multinex and Multinex-Nano models, with their compact parameter counts (45KB and 20KB respectively), facilitates broader adoption and further research (albrateanu/multinex, March 2026). The project page for Multinex is also live, offering additional resources and demonstrations of the technology. This public release strategy from the University of Manchester (manchester.ac.uk, June 2026) aims to accelerate advancements in low-light image enhancement and potentially inspire similar models that prioritize efficiency without compromising performance.
Read full article at bioengineer.org
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source