A deep-learned pose optimizer framework that uses spatio-temporal architecture to learn postural information.
Mahir Patel
665 Commonwealth Ave, Boston, MA 02215
mahirp@bu.edu
Third-Year Computer Vision Ph.D. student supervised by Prof. Margrit Betke, Prof. Bryan Plummer, and Prof. Michael Hasselmo. I am interested in efficiently interpreting the environment from different modalities. Outside of academia, I enjoy listening / playing heavy metal, horror movies, and video games!
Research Projects
A multi-view multi-modal Rodent dataset
An automated tool to analyze One-Hybrid yeast assays to accelerate research.
A visualization software counter-part for the D.I.S.H.A package.
Publications
Peer-Reviewed Publications
Coding of Space and Time for Memory Function
Space, Time, and Memory, eds. Lynn Nadel & Sara Aronowitz
(Oxford University Press)
How does the response of egocentric boundary cells depend upon the coordinate system of environmental features?
bioRxiv
ExeChecker: Where Did I Go Wrong?
European Conference on Computer Vision Workshops (ECCVW)
Paired yeast one-hybrid assays to detect DNA-binding cooperativity and antagonism across transcription factors
Nature Communications
Animal Pose Tracking: 3D Multimodal Dataset and Token-based Pose Optimization
International Journal of Computer Vision (IJCV)
Thesis
Learned Structural And Temporal Context For Dynamic 3D Pose Optimization And Tracking
Boston University
Abstracts & Posters
BU-CVKit: Extendable Computer Vision Framework for Behavioral Neuroscience
Society For Neuroscience
Longitudinal tracking of spatial responses to environmental manipulations in the retro-splenial cortex
Society for Neuroscience
BU-CVKit: Extendable Computer Vision Framework for Species Independent Tracking and Analysis
CV4Animals@CVPR
Animal Pose Tracking: 3D Multimodal Dataset and Token-based Pose Optimization
CV4Animals@CVPR
A Rate Code of Environmental Alteration and High-Resolution 3D Pose Tracking During Behavior
Boston University
Neural correlates of environmental alterations in retro-splenial cortex and high-resolution infrared tracking of behavior
Professional Experience
- Developed video capabilities for Dragon Copilot, building audio-video summarization pipelines grounded in video scene graphs.
- Presented research outcomes to senior leadership and product teams across Microsoft.
- Partnered with clinicians to gather feedback and refine product direction.
- Setting up multi-view tracking cameras and hardware synchronization.
- Writing recording software and neural data analysis/visualizing tools.
- Designing new architecture for end-to-end 3D pose estimation and optimization.
Academic Experience
- Implemented 3D reconstruction from multi-view thermal cameras and designed configurable software to generate analytical data such as displacement vectors and velocity vectors.
- Customized the 3D calibration cube with Peltier devices for thermal cameras.
- Annotated 100s of frames for each view and trained deep neural network to generate landmarks on rodents.
- Ported old proprietary FLIR SFMOV video editing code to Linux Systems.
- Scraped more than 15,000 research articles from PubMed, from different reputed journals.
- Created a software to help annotate abstracts and build the dataset.
- Fine tuned PubMedBERT by Microsoft to classify research articles.
- Designed a computer vision pipeline to localize and segment quad-colonies from Yeast one-hybrid assay images, assigning scores to represent the interactions of different permutation of Transcription Factors, to accelerate the research in fields like Cancer research and immune responses.
Education
Boston University
Computer Vision
Boston University
Master of Science
Artificial Intelligence
Current GPA: 3.82/4.0
Gujarat Technological University
Bachelor of Engineering
Computer Engineering
CGPA: 8.94/10.0