Summary 🧬

Research paper accepted at ESANN 2022 that benchmarks deep segmentation backbones for renal glomeruli detection on mouse histology.

The journey 🧭

• The story begins with my undergraduate thesis, where I leveraged DeepLab V2 to segment renal sections of mice.
• For this follow-up, I introduced MobileNet to both save computation and compare performance trade-offs between heavy and lite encoders.
• The analysis focused on precision vs. speed, exploring how lightweight encoders can maintain accuracy while enabling faster inference for eventual deployment in lab gadgets.

Outcomes & takeaways 📈

• The paper is publicly available, with the full method and results shared at the conference link.
• Code and trained models are published on GitHub, making it easy to reproduce or extend the segmentation pipeline.
• The project reaffirmed my interest in biomedical applications of deep learning and taught me how to document results for a scientific audience.

Resources 🔗

• ESANN 2022 paper
• KerasSegmentation repository
• Bachelor thesis (precursor)