Related Experiment Video
Updated: Jun 3, 2026
07:12
Whole-cell Super-Resolution Imaging via DNA-PAINT on a Spinning Disk Confocal with Optical Photon Reassignment
Published on: January 6, 2026
Overcoming resolution constraints in automated colony counting via a high-performance deep learning framework using
Sercan Külcü1, Duygu Balpetek Külcü2
1Computer Engineering Department, Giresun University, 28200, Giresun, Turkey. sercan.kulcu@giresun.edu.tr.
Scientific Reports
|June 1, 2026
Summary
Tiling high-resolution images improves lightweight AI models for bacterial colony counting. This method significantly boosts accuracy, making it ideal for lab use.
Area of Science:
- Computer Vision
- Microbiology
- Machine Learning
Background:
- Lightweight AI models struggle with bacterial colony counting on downscaled high-resolution images.
- Downscaling to 640×640 pixels degrades colony details, impacting counting accuracy.
Purpose of the Study:
- To enhance the performance of lightweight AI models in bacterial colony counting.
- To address the limitations of image downscaling in high-resolution Petri dish analysis.
Main Methods:
- Implemented a tiled training and SAHI-based tiled inference pipeline.
- Divided full-resolution images into overlapping 640×640 tiles, preserving native resolution.
- Evaluated three nano-scale YOLO models (YOLOv5n, YOLOv8n, YOLOv11n) on a 24-class bacterial colony dataset.
Main Results:
- The tiled approach achieved 95.4-96.9% mAP@0.5, a significant improvement over the conventional resize method (44.9-66.3%).
- YOLOv11n with tiling reached 96.9% mAP@0.5 and 58.2% mAP@0.5:0.95 with only 2.6 million parameters.
- Inference time was under 320 ms on a laptop RTX 3050 GPU, demonstrating efficiency.
Conclusions:
- Tiled inference with moderate overlap dramatically outperforms image resizing for small, densely packed objects like bacterial colonies.
- The proposed lightweight pipeline offers a robust and efficient solution for routine laboratory bacterial colony counting.
