Arnav Shah
Pronouns: he/him
Research Mentor(s): Ada Eban-Rothschild
Research Mentor School/College/Department: Psychology / LSA
Program:
Authors: Arnav Shah, Gary Ciptura, Gaurav Kaul, Ada Eban-Rothschild
Session: Session 1: 9:00 am – 9:50 am
Poster: 49
Abstract
Animal behavior identification is a critical element in behavioral studies across various fields. However, monitoring animal footage and analyzing it can be a challenging and time-consuming task for researchers. To overcome this challenge, we propose a new toolkit that can identify behavioral traits of animals in any lab environment via video feed. Our toolkit will utilize the SlowFast network by FAIR, which automates the task of behavioral classification using lab video data. We created a diverse dataset with varied video lengths (5-30 seconds) and will use the proposed DAMM detector to identify mice in these environments. Next, we will feed these videos into the SlowFast network using a ResNet backbone through two pipelines: a slow pipeline that works on low framerates to identify spatial semantics and a fast pipeline to detect temporal dynamics. We will then fine-tune our model by using supervised learning to increase accuracy. Our toolkit’s performance will be tested on a separate dataset, confirming its accuracy and reliability in automated behavioral labeling. By integrating the SlowFast network, DAMM Detector, and unsupervised learning, our toolkit will offer a robust solution that enables researchers to identify behaviors on different animals with increased efficiency and accuracy in behavioral analysis via our generated API with the hopes of having a better understanding in these studies.
