Isabelle Baker
Pronouns: she/her
Research Mentor(s): Bing Ye
Research Mentor School/College/Department: Life Sciences Institute, University of Michigan / Medicine
Program: UROPF
Session: Session 1 (9:00am – 9:50am)
Authors: Isabelle Baker, Yujia Hu, Bing Ye
Abstract
Ye lab has developed a computational tool called LabGym for efficiently and accurately analyzing user-defined behaviors across animal species. However, LabGym has several limitations preventing it from being adopted more broadly. The most significant limitation of the current version of LabGym is that it requires the illumination and the background in videos to be static over time. Thus, if the background is non-static, LabGym struggles to identify the background and therefore fails to remove it properly, which restricts its application in videos with a dynamic background. The goal of this research is to develop a new detection algorithm that can be integrated into LabGym, which will broaden the application of this program. To accomplish this goal, a detailed understanding was acquired of how the current version of LabGym approached the problem and then solutions were sought. After reading academic articles and comparing different alternative methodologies, it was concluded that the identification and segmentation of animals from the background, rather than the removal of the background, would be the best approach. For this purpose, a new animal segmentation program was developed that is completely different from the detection module in the current LabGym, based on Detectron2, a framework developed by Meta’s artificial intelligence team. The new animal segmentation program was integrated into LabGym and an easy-to-use graphical user interface was developed to facilitate the usage of the program without requiring computer programming skills. With all these new features, LabGym has been completely upgraded to LabGym 2.0, which is now applicable to behavioral analysis without restrictions on the video recording settings, environment, behavior types, or animal species. The new LabGym will enable more scientists to be able to use the tool for their research, likely in ways we have yet to even consider.
