Dr. Sana Alamgeer (She/Her)
PhD | Electronic and Automation Systems Engineering
About Me
I was born in Pakistan, where I completed my higher education, earning a master’s degree in computer science from Bahaudding Zakariya University, Multan, Pakistan, and a Master of Philosophy in computer science from Air University, Multan Campus, Pakistan. Later, I moved to Brazil for my doctoral studies at the University of BrasĂlia, Brazil, where I developed deep learning-based methods for visual quality assessment of 2D and 4D images and videos, and visual attention models for 360-degree videos.
During my stay in Brazil, I was fortunate to work on a project that allowed me to develop natural language processing (NLP) tools, including a synonym prediction system for Portuguese. After completing my Ph.D., I served as a Senior Data Science Coach at CeADAR in Ireland, where I designed and delivered interactive AI modules for business executives, focusing on AI-relatd topics such as Demystifying AI, Generative AI, Prompt Engineering. I also developed an interactive online module delivery system, enabling engaging and effective learning experiences for learners.
Currently, I am a Postdoctoral Researcher at Texas State University, focusing on computer vision and time-series analysis for healthcare applications. I have my YouTube channel "Coderific", where I publish videos on programming. Considering that every problem has multiple solutions, you will find one of those solutions in this channel. Through this platform, I aim to help research students doing small tasks so that their big projects do not stop.
My research interests focus on computer vision, generative AI, time-series analysis, prompt engineering, synthetic time series data generation, and interpretable machine learning, with applications in healthcare and multimedia.
Publications
Journals
A survey on visual quality assessment methods for light fields
Sana Alamgeer and Mylène C. Q. Farias
Signal Processing: Image Communication - ELSEVIER (2022)
Blind visual quality assessment of light field images based on distortion maps - [code]
Sana Alamgeer and Mylène C. Q. Farias
Frontiers Signal Processing – Image Processing (2022)
A two-stream CNN-based visual quality assessment method for light field images - [code]
Sana Alamgeer and Mylène C. Q. Farias.
Journal of Multimedia Tools and Applications (2022)
CNN-based no-reference video quality assessment method using a spatiotemporal saliency patch selection procedure - [code]
Sana Alamgeer, Muhammad Irshad, and Mylène C. Q. Farias.
Journal of Electronic Imaging - SPIE (2021)
Conferences
Using a Diverse Neural Network to Predict the Quality of Light Field Images - [code]
Sana Alamgeer, André H. M. Costa and Mylène C.Q. Farias
IEEE Workshop on Multimedia Signal Processing (2023)
Light Field Image Quality Assessment with Dense Atrous Convolutions - [code]
Sana Alamgeer and Mylène C. Q. Farias
IEEE International Conference on Image Processing - ICIP (2022)
360RAT: A Tool for Annotating Regions of Interest in 360-degree Videos
Lucas Althoff, Myllena Prado, Sana Alamgeer, Alessandro Rodrigues e Silva, Ravi Prakash, Marcelo M Carvalho, and Mylène C. Q. Farias
WebMedia '22: Proceedings of the Brazilian Symposium on Multimedia and the Web (2022)
Deep Learning-Based Light Field Image Quality Assessment Using Frequency Domain Inputs - [code]
Sana Alamgeer and Mylène C. Q. Farias
14th International Conference on Quality of Multimedia Experience - QoMEX (2022)
Long Short-Term Memory based Quality Assessment of Light Field Images - [code]
Sana Alamgeer and Mylène C. Q. Farias
IEEE International Conference on Multimedia and Expo Workshops - ICME (2022)
Sana Alamgeer, Muhammad Irshad, and Mylène C. Q. Farias
MMSys '22: Proceedings of the 13th ACM Multimedia Systems Conference (2022)
Muhammad Irshad, Camilo Sanchez-Ferreira, Sana Alamgeer, Carlos H. Llanos, and Mylène C. Q. Farias
Conference of IS&T International Symposium on Electronic Imaging (2021)
Perceptual quality assessment of enhanced images using a crowd-sourcing framework
Muhammad Irshad, Alessandro Silva, Sana Alamgeer, and Mylène C. Q. Farias
Conference of IS&T International Symposium on Electronic Imaging (2020)
Blind image quality assessment based on multiscale salient local binary patterns
Pedro Garcia Freitas, Sana Alamgeer, Welington Akamine, and Mylène C. Q. Farias
MMSys '18: Proceedings of the 9th ACM Multimedia Systems Conference (2018)
Projects
Time Series Analysis
Currently, as a Postdoctoral Researcher at Texas State University, I am working on time-series analysis projects with a focus on healthcare applications. My research involves developing efficient algorithms for critical tasks such as fall detection in elderly patients and seizure prediction using EEG data. These projects integrate explainable AI techniques to enhance model interpretability, ensuring transparency and trustworthiness in decision-making processes. Additionally, I explore domain adaptation strategies to improve model robustness across diverse data sources, aiming to create reliable, real-time solutions for healthcare monitoring and patient safety.
Generative AI
In the domain of Generative AI, my work addresses data scarcity in healthcare by developing advanced models for generating realistic time-series and motion data. Leveraging advanced methods like diffusion models large language models (LLMs), we create synthetic fall motion data and improve predictive model performance.
Knowledge Distillation
This undergoing project focuses on developing Lightweight Human Activity Recognition methods, optimized for resource-limited wearable devices like smartwatches. This framework leverages knowledge distillation to transfer insights from a multimodal teacher model to a lightweight student model that uses only accelerometer data. Additionally, we explore various knowledge distillation techniques, including both forward and reverse Kullback-Leibler Divergence (KLD), to enhance the knowledge transfer process.
Digital Twins for Human Activity Recognition
This is an undergoing project that focuses on developing digital twins, sophisticated, physics-based models replicating an individual’s unique physiological and biomechanical traits. These models simulate motion dynamics, enabling the analysis of gait patterns and fall scenarios by introducing variables like sudden weight shifts or directional changes.
Natural Language Processing
As an NLP enthusiast, I worked on a project where I analyzed pre-trained models on the Portuguese language to build an algorithm capable of predicting synonyms for every word. With a combination of two successful pre-trained models, I was able to achieve accurate predictions without the need to train a network from scratch. This project showcases my expertise in NLP, particularly in the Portuguese language, and highlights my ability to creatively leverage existing resources to deliver high-quality results.
Computer Vision | Github
In my recent project, I designed a novel model aimed at accurately predicting regions of interest (ROIs) in 360â—¦ videos. The ROI is crucial for various applications such as predicting view-ports, optimizing video cuts for live streaming, and improving the overall viewing experience. The project involved training and testing a hybrid saliency model that was developed to identify the saliency regions representing the ROIs. The methodology comprised of preprocessing video frames, predicting ROIs using the hybrid saliency model, and post-processing to obtain the final output. The performance of the proposed model was then compared with the subjective annotations of the 360RAT dataset, showcasing my expertise in developing innovative solutions in the field of computer vision.
Resources
Programming Tools & Libraries:
Python
Quality Assessment Databases:
Events
Upcoming Events:
no upcoming events yet.