- Updated: August 25, 2026
- 2 min read
MicroAUNet: Boundary‑Enhanced Multi‑scale Fusion with Knowledge Distillation for Colonoscopy Polyp Segmentation
MicroAUNet: A Lightweight, High‑Performance Solution for Real‑Time Polyp Segmentation
Colorectal cancer remains a leading cause of mortality worldwide. Early detection and accurate segmentation of polyps during colonoscopy are essential for reducing this burden. While many deep‑learning models achieve high accuracy, they often suffer from heavy computational demands, limiting their use in real‑time clinical settings.
In this article we present MicroAUNet, a lightweight attention‑based segmentation network designed specifically for colonoscopy polyp image segmentation. MicroAUNet combines depthwise‑separable dilated convolutions with a single‑path, parameter‑shared channel‑spatial attention block, effectively strengthening multi‑scale boundary features while keeping the model size minimal.
Key innovations include:
- Boundary‑enhanced multi‑scale fusion that captures fine‑grained polyp margins.
- Progressive two‑stage knowledge distillation that transfers both semantic and boundary cues from a high‑capacity teacher model.
- Extremely low model complexity (under 1 M parameters) enabling real‑time inference on standard endoscopic hardware.
Extensive experiments on public benchmarks demonstrate state‑of‑the‑art accuracy with a fraction of the computational cost, making MicroAUNet a practical choice for clinical deployment.
For a detailed technical description, source code, and pretrained models, visit the official repository. Additional resources and related projects are available on our site:
Stay tuned for upcoming tutorials and integration guides.
Andrii Bidochko
CTO UBOS
Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.