Advancing computer vision and edge AI through practical research
University of Science and Technology of China
This research explores the adaptation of Segment Anything Model (SAM) for real-time fire segmentation on edge devices. The work bridges the gap between large-scale vision foundation models and lightweight architectures suitable for resource-constrained environments, with a focus on practical deployment scenarios.
2026 9th International Conference on Image and Graphics Processing (ICIGP '26), Wuhan, China
Accepted
@inproceedings{profsam2025,
author = {Emmanuel U. Ugwu and Xinming Zhang},
title = {Promptable Fire Segmentation: Unleashing SAM2's Potential for Real-Time Mobile Deployment with Strategic Bounding Box Guidance},
booktitle = {2026 9th International Conference on Image and Graphics Processing (ICIGP '26)},
year = {2026},
address = {Wuhan, China},
month = jan,
note = {to appear}
}
The Visual Computer
Under Review
@article{Ugwu2025LimFUNet,
title = {Enhancing Real-time Fire Segmentation: LimFUNet with SE-Enhanced Ghost Convolutions for Edge Computing Applications},
author = {Ugwu, Emmanuel U. and Zhang, Xinming and Tesfay, Semere G. and Mehmood, Muhammad Hamza},
journal = {The Visual Computer},
year = {2025}
}
Machine Vision and Applications
Under Review
@article{Ugwu2025EdgeAware,
title = {Edge-Aware Dual Path Network for Medical Image Classification},
author = {Muhammad Hamza Mehmood, Xinming Zhang, and Emmanuel U. Ugwu},
journal = {Machine Vision and Applications},
year = {2025}
}
Adapting large-scale vision models (SAM, SAM2) for specific tasks through prompt engineering and efficient fine-tuning strategies.
Developing lightweight CNN architectures using ghost convolutions, squeeze-and-excitation blocks, and knowledge distillation.
Designing efficient segmentation networks for fire detection, medical imaging, and environmental monitoring applications.
Optimizing deep learning models for mobile and embedded devices with limited computational resources.