An interactive class-agnostic detection framework for generalizable colony counting
5 affiliations
- 1State Key Laboratory of Integrated Chips and Systems, College of Integrated Circuits and Micro-Nano Electronics, Fudan University
- 2Shanghai Institute of Infectious Disease and Biosecurity, School of Basic Medical Sciences, Fudan University
- 3Shanghai Municipal Center for Disease Control and Prevention
- 4Lingnan University
- 5Shanghai Turtle Technology Co. Ltd.
Abstract
Introduction: Microbiological monitoring is an essential component of laboratory and industrial quality-control workflows, yet accurate colony counting remains challenging to generalize across diverse colony types and imaging conditions. To address this challenge, we propose the Interactive Colony Counter (ICC), a lightweight and class-agnostic detection framework for microbial colony counting.
Methods: Aiming to improve generalization across diverse conditions, ICC takes the entire colony plate image as input and iteratively integrates user guidance, in the form of bounding boxes for target colonies and corrections of detection errors. ICC consists of three components: the Cross Stage Partial Fusion Network (CSPFNet) as the backbone, the Contrastive-Guidance Encoder (CGE) for processing user guidance, and the Multi-scale Detection Module (MDM) as the detection head. By training on the AGAR dataset with a class-agnostic strategy, ICC learns general colony features rather than relying on type-specific labels.
Results: Under simulated user guidance, experimental results show that ICC achieves competitive performance on the AGAR dataset, with a mean Average Precision (mAP) at IoU 0.5:0.95 of 60.7 and 62.9% at different resolutions. To assess generalization beyond the training domain, ICC is further evaluated on the ADBC dataset under different imaging conditions, which also includes species not represented in the training data. In addition, we demonstrate the feasibility of ICC-assisted annotation by constructing the B. lactis -Colony dataset for colony detection, which is generated from Bifidobacterium lactis and constructed using the ICC-assisted annotation workflow.
Discussion: Overall, ICC demonstrates strong potential for colony detection and dataset annotation, providing an effective lightweight framework with iterative user guidance for microbiological imaging tasks.
Related
- Auditable Algebraic Counting Field for Cryptic-Pocket Detection from Apo Structures
- An open structure-guided RNA foundation model for generalizable prediction
- A generalizable Hi-C foundation model for chromatin architecture, single-cell and multiomics analysis across species
- StabilityArc: Decoding Protein Sequence Embeddings into Generalizable Stability Landscapes
- SpotMAX: A generalist framework for multidimensional automatic spot detection and quantification
BibTeX
@article{fan2026interactive,
title = {{An interactive class-agnostic detection framework for generalizable colony counting}},
author = {Jingxing Fan and Jinrong Shen and Yanqi Zhu and Pangen Hu and Haofeng Li and Gangwei Xu and Zhehao Zhao and Dongping Wu},
journal = {Frontiers in Microbiology},
year = {2026},
month = oct,
doi = {10.3389/fmicb.2026.1967912},
url = {https://www.frontiersin.org/articles/10.3389/fmicb.2026.1967912/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - An interactive class-agnostic detection framework for generalizable colony counting
AU - Jingxing Fan
AU - Jinrong Shen
AU - Yanqi Zhu
AU - Pangen Hu
AU - Haofeng Li
AU - Gangwei Xu
AU - Zhehao Zhao
AU - Dongping Wu
PY - 2026
DA - 2026/10/21
JO - Frontiers in Microbiology
DO - 10.3389/fmicb.2026.1967912
UR - https://www.frontiersin.org/articles/10.3389/fmicb.2026.1967912/full
AB - Introduction: Microbiological monitoring is an essential component of laboratory and industrial quality-control workflows, yet accurate colony counting remains challenging to generalize across diverse colony types and imaging conditions. To address this challenge, we propose the Interactive Colony Counter (ICC), a lightweight and class-agnostic detection framework for microbial colony counting. Methods: Aiming to improve generalization across diverse conditions, ICC takes the entire colony plate image as input and iteratively integrates user guidance, in the form of bounding boxes for target colonies and corrections of detection errors. ICC consists of three components: the Cross Stage Partial Fusion Network (CSPFNet) as the backbone, the Contrastive-Guidance Encoder (CGE) for processing user guidance, and the Multi-scale Detection Module (MDM) as the detection head. By training on the AGAR dataset with a class-agnostic strategy, ICC learns general colony features rather than relying on type-specific labels. Results: Under simulated user guidance, experimental results show that ICC achieves competitive performance on the AGAR dataset, with a mean Average Precision (mAP) at IoU 0.5:0.95 of 60.7 and 62.9% at different resolutions. To assess generalization beyond the training domain, ICC is further evaluated on the ADBC dataset under different imaging conditions, which also includes species not represented in the training data. In addition, we demonstrate the feasibility of ICC-assisted annotation by constructing the B. lactis -Colony dataset for colony detection, which is generated from Bifidobacterium lactis and constructed using the ICC-assisted annotation workflow. Discussion: Overall, ICC demonstrates strong potential for colony detection and dataset annotation, providing an effective lightweight framework with iterative user guidance for microbiological imaging tasks.
SN - 1664-302X
ER -