[4O4-IS-2b] Machine learning

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)

[4O4-IS-2b-04] Performance Evaluation of the Animal Tracking Framework of ilastik on Multi-Species Fish Datasets

〇Raj Rajeshwar Malinda¹ (1. University of Hyogo, Japan)

regular,[[online]]

Keywords:

Bio-complexity,ilastik,Machine learning,Object detection,Tracking

Multi-object detection and tracking tasks are complex, labor-intensive, and often considered challenging across disciplines ranging from biomedical to ecological sciences. Although a wide variety of tools and techniques are available for domain-specific annotated datasets, cross-domain performance evaluation of these tools is still often overlooked. Therefore, this study aims to provide a practical performance evaluation of the animal tracking framework of ilastik, a machine learning-based interactive image analysis application, primarily focused on (bio)imaging datasets, using publicly available open-source multi-species fish datasets. These datasets were selected to encompass various parameters, such as number of objects, non-homogeneous background complexity, and image quality. The performance evaluation demonstrates an easy-to-implement pipeline with minimal coding requirements for processing within the animal tracking framework, to produce high-quality segmented probability maps with a small number of training data, and accuracy shows reliable performance. In conclusion, this study offers an insightful practical and qualitative cross-disciplinary evaluation of the ilastik tracking framework, highlighting its intuitive tracking effectiveness for applications in computer vision, automation, and bio-complexity research.