Cichlid Computer Vision Project
Researchers: Charles Clark, Kailey Quesada, Thuan Nguyen, Adam Thomas, Jeanette Schofield
Collaborator: Dr. Patrick McGrath
The project will focus on analyzing the newly available data from multiple cichlid species using the Cichlid Bower Tracking Repository. This will involve processing the raw video footage through the existing pipeline to extract annotated behavioral data. Concurrently, a multi-species animal tracking dataset will be curated by combining data from the various species. Building upon the initial analysis, the project will explore data distillation techniques to improve the efficiency and scalability of the tracking process. This may involve techniques such as data subsampling, compression, or feature extraction to reduce the computational overhead without significantly compromising accuracy. Novel challenges such as occlusion, where animals partially or fully obscure each other, will be tackled through the development of specialized algorithms and model architectures.
Publications:
- Johnson, Z.V., Arrojwala, M.T.S., Aljapur, V. et al. Automated measurement of long-term bower behaviors in Lake Malawi cichlids using depth sensing and action recognition. Sci Rep 10, 20573 (2020). https://doi.org/10.1038/s41598-020-77549-2
- Long, Lijiang & Johnson, Zachary & Li, Junyu & Lancaster, Tucker & Aljapur, Vineeth & Streelman, Jeffrey & Mcgrath, Patrick. (2020). Automatic Classification of Cichlid Behaviors Using 3D Convolutional Residual Networks. iScience. 23. 10.1016/j.isci.2020.101591.
Natural Florida History Museum Project
Researchers: Thomas Deatherage, Vy Nguyen, Romouald Dombrovski
Collaborator: Dr. Arthur Porto
Natural history collections are invaluable resources for scientific research, education, and public engagement. However, the sheer volume and diversity of specimens often make it challenging to efficiently search and retrieve specific items. The goal of this project is to develop a sophisticated search interface that leverages advanced machine learning techniques to embed images from natural history collections, enabling users to search the database using images or natural language queries.
Researchers: Thomas Orth, Michael Bock
Collaborator: Charlotte Alexander, J.D.
One of this project’s goals is to summarize legal documents from the clearinghouse.net to automate Clearinghouse’s work with law students. Another target that the team is working on is a document classification project on a dno case dataset from UPenn. The team attempts to automatically populate certain fields such as the allegation, the companies involved, etc.
Researchers: Karol Gutierrez Suarez, Alejandro Gomez, Víctor Fernandez
Collaborator: Charlotte Alexander, J.D.
This project is to analyze the use of NLP on court case documents from the Dominican Republic to estimate processing duration in order to optimize case triaging. The narrowest goal is to extract from each sentencia the procedural history of the case, focusing on dates. From that, a timeline of each case can be constructed. Other variables can also be built, e.g. the type of the case, the court, the judge, etc. and eventually a model can be built that helps identifying or even predicting the types of cases that take the longest to progress from step A to the final outcome. The secondary goal is to help the public policy people in the judiciary identify possible interventions that would help with delays. The next-level goal is to demonstrate to the judiciary the value of structured data. If the judges realize that structuring their sentencias more consistently will make this kind of analysis easier, and therefore make their courts more efficient, then they might be motivated to adopt and follow some rules about standardizing the information that they include in the sentencias. The goal would ultimately be to increase standardization to make information extraction easier. An additional next-level goal is to use this work as proof of concept for other countries’ courts/ legal systems, as a way to demonstrate the value of a data science approach to courts and court operations.
Researchers: Mercedes Quintana, Philip Woolley, Jacob Dallaire, Ruiqing Wang, Ayush Parikh
Collaborator: Dr. James Stroud
The project is in collaboration with the Stroud lab in the Department of Biological Sciences, within the domain of Ecology and Evolution. The research question is to investigate the morphology to fitness connection in adaptive evolution, and specifically, the performance cost to missing a leg in lizards. Given videos of experimental trials involving lizards jumping and sprinting, the team uses DeepLabCut (DLC) to track the positions of key body parts, from which biophysical information of interest to biologists can be extracted.
Researchers: Ming Zhong, Philip Woolley, Shuyu Tian
Collaborators: Dr. James Stroud, Prakhar Kaushik
The Lizard Jaw Segmentation project aims to automate the segmentation of teeth and lower jaws of Anolis lizards’ 3D Micro CT scans. Traditional segmentation of the scans requires manual efforts using custom software, which can take up to two hours per scan. Automating the process will significantly reduce process time, which will allow researchers to conduct larger-scale data collection and comparative studies across not just different Anolis lizards, but also different species.
Researchers: Wen Han Chia
Collaborators: Dr. James Stroud, llia Jahanshahi
The Florida Anole Species Classification project aims to develop a robust classification pipeline for identifying five common Anolis species from photographs, primarily to support a community science initiative with middle school students in Miami. Building upon an extensive dataset of over 80,000 verified iNaturalist photographs, this project seeks to improve the current classification system, which, despite having access to substantial training data, currently achieves only 35% accuracy (compared to a random baseline of 20%). The development of this classification pipeline will serve as the foundation for a broader educational tool, whether implemented as a mobile application or web platform, that enables students to receive immediate probability-based species identification feedback before submitting their observations to iNaturalist, thereby enhancing the quality of citizen science data collection while engaging young students in herpetological research.
Researchers: Taran Lau
Collaborators: Dr. James Stroud, Danil Akhtiamoiv, Marium Yousuf
The Lizard Locomotion Analysis project aims to develop a comprehensive understanding of lizard running behavior through advanced computer vision techniques, specifically utilizing DeepLabCut and SLEAP for precise pose estimation and movement analysis. By creating detailed anatomical landmarks across the lizard’s body, this project will enable quantitative analysis of various locomotor patterns and behavioral characteristics, moving beyond traditional center-of-mass tracking to capture complex biomechanical interactions during movement. Through this open-source approach, we will document multiple behavioral patterns across different body segments, contributing to a broader understanding of lizard locomotion while establishing standardized protocols for future research in animal biomechanics. This work will not only advance our understanding of lizard movement patterns but also contribute to the growing field of automated behavioral analysis in biological research
Lizard LiDAR Vegetation Analysis
Researchers: Amir Hossein Alikhah Mishamandani, Fan Yang, John Hagood
Collaborators: Dr. James Stroud, Jefferey Cannon, Dory Peters, Baidik Chandra
The project aims to develop innovative algorithms for processing and analyzing terrestrial LiDAR (Light Detection and Ranging) scans of natural vegetation, with the goal of creating comprehensive, open-source software packages in Python or R. This project will focus on creating novel computational methods to extract, process, and analyze complex vegetation structure data from LiDAR point clouds, enabling more accurate and automated assessment of natural vegetation characteristics. We are especially interested in segmenting and measuring elements of branch and vegetation structure. By developing these algorithms into accessible software packages, the project will provide the scientific community with robust tools for vegetation analysis, supporting applications in ecology, forestry, and environmental monitoring.
Researchers: Yoon Kim, Michael Falter, Vikas Agarwal, Eve Dang, Anthony Trevino
Collaborators: Bree Shi, Professor Lytle
The goal of this project is to create and maintain a structure for large research groups in higher education. Code base solutions, contribution tracking, resource management, researcher support, and program development are included in this project. Additionally, members of this team often participate in additional projects individually, as directed.
Researchers: Nikita Angarski, Omar Moursy, Steve Foryoung
Collaborators: Dr. Arthur Porto, Kseniia Shilova, llia Jahanshahi
This project aims to develop a more biologically informed approach for registering 3D models of anatomical structures, such as mouse skulls. Traditional registration methods (e.g., Coherent Point Drift) treat point clouds in a purely geometric way and often ignore the underlying biological constraints of how real organisms vary in shape. By incorporating statistical shape models—built from actual biological data—into the registration pipeline, the team seeks to achieve more accurate correspondences between specimens. This improvement will ultimately enhance automated measurements, morphological analyses, and potentially benefit broader applications where 3D biological data play a role.